A hybrid self driving ground vehicle and a method thereof

Abstract

The present invention relates to hybrid self-driving ground vehicle (100) for autonomous navigation In one embodiment, the hybrid self-driving ground vehicle (100) comprises a mobility platform (102) configured to actuate movement of the vehicle, an odometry sensor system (104) configured to continuously estimate translational and rotational motion of the mobility platform, a visual sensing system (106) comprising plurality of two-dimensional camera configured to acquire image data of an environment, an obstacle detection system (108) configured to detect dynamic and static obstacles, one or more processors (110, 112) operatively coupled to the odometry sensor system, the visual sensing system, and the obstacle detection system, wherein the one or more processors configured to: select between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode, continuously localize the vehicle using the odometry sensor system, intermittently perform environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment and autonomously navigate the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

Description

A HYBRID SELF DRIVING GROUND VEHICLE AND A METHOD THEREOFTECHNICAL FIELD

[0001] The present invention relates to the field of autonomous vehicles and more particularly to the autonomous vehicle for transportation which uses simplified 2D mapping, enabling higher energy saving for longer battery life and making it easier for users to understand and interact with the system.BACKGROUND OF THE INVENTION

[0002] Automated material movement is a critical component of industrial automation, particularly within the context of Industry 4.0. Traditionally, large factories have employed Automated Guided Vehicles (AGVs) that run on tracks fixed to a floor. More recently, there has been a shift towards using Autonomous Mobile Robots (AMRs). AMRs leverage 3D LiDAR technology to map their surroundings and plan their path dynamically and do not need fixed tracks. Additionally, other types of mobile robots, utilize 2D markers placed on the floor at predefined distance to form a grid like structure.

[0003] Automated Guided Vehicles (AGVs) is a well-known material handling system, and the technology used for navigation is following fixed paths created by metal tapes, magnetic tapes wherein the sensors are Metal sensors, RFID sensors, infrared sensors, magnetic sensors. However, the problem faced in the Automated Guided Vehicle is limited flexibility and adaptability, requiring physical changes to the environment for route adjustments, need for re -programming every time new path is added, dependence on the maintenance of tapes, potential operational failure due to deterioration of the tapes with floor cleaning and other activities and limitation to indoor usage.

[0004] Autonomous Mobile Robots (AMRs) is another material handling system, and the technology used for navigation is SLAM (Simultaneous Localization and Mapping) based on sensors like 3D Lidar or 3D Camera. However, the problem faced in the Autonomous Mobile Robots (AMRs) is high cost and complexity, potential issues with haphazard movements in highly vehicle dense environments, loss of efficiency and accuracy in dynamic environments, atypical scenario in factories.

[0005] Marker-Based Mobile Robot is another material handling system, and the technology used for navigation follows a grid of encoded markers pasted at a predefineddistance from each other, on the floor of a facility. The sensors used in Marker-Based Mobile Robots are Laser scanners and cameras. However, the problem faced with such Robots is that the number of markers increase by power of 2 with increase in floor area. For example, if markers are placed 1 meter apart, then for 3 sqm area, typically 3A2=9 markers are needed, and for 100 sqm area, typically 100A2= 10000 markers are needed. In addition, there is dependence of navigation accuracy on the accuracy in pasting the markers, their maintenance and potential interference with floor cleaning and other activities, limiting flexibility, adaptability, only indoor usage and potential failures in case of inaccurate pasting of markers.

[0006] Further limitation of Automated Guided Vehicles (AGVs) is that they depend on the physical tracks. Generally, AGVs require the installation of physical tracks such as tapes, sensors, or markers on the work-floor, which results in downtime whenever there is new installation. Since AGVs follow pre-programmed paths, any change in layout or route requires the system to be reprogrammed. This not only involves additional costs but also demands technical staff with programming skills making them non- flexible. Besides, these tracks can be damaged by routine cleaning or wear and tear, leading to frequent maintenance needs and higher operational downtime. Therefore, it lacks efficiency. AGVs have limited autonomy and are incapable of responding intelligently to changes in the environment and dynamic obstacles like human beings, making them suitable only for one specific application and therefore, non-adaptable.

[0007] Further limitation of Autonomous Mobile Robots (AMRs) is that they depend on High Data and Power Requirements. AMRs typically use 3D LiDARs or 3D cameras to generate 3D point cloud map for navigation, which are difficult to understand by a human eye therefore, inconvenient to use by robot operators. Also, these require continuous processing of Lidar or image frames to maintain positional accuracy while in motion or stationery, creating close to a million data points per second requiring a high bandwidth of close to 80Mbps per robot and high processing power requirement of close to 180Watts which drains the battery very fast. Dynamic environments where the physical layout or the placement of objects frequently changes, like factories, warehouses and outdoors, AMRs struggles to maintain reliable navigation accuracy. This makes AMRs in-efficient even though it is more expensive.

[0008] Further, some AMRs are restricted to indoor use due to their reliance on floor markers, and those equipped to handle outdoor environments must contend with variable lighting, weather conditions, and other unpredictable elements that can impede their functionality. To improve overall reliability, many AMRs use both 3D cameras and 3D LiDAR for navigation. This increases the cost and complexity of AMRs, making them less accessible for smaller operations or those with limited budgets. The increased data processing requirement also adds to the operational overhead.

[0009] Furthermore, integrating the above-mentioned automated systems (AGVs and AMRs) into existing workflows can be challenging, especially in facilities that are not originally designed with automation in mind. Retrofitting these systems can be costly and disruptive. AMRs may operate in environments potentially shared with humans. To maintain safety, AMR stops at every obstacle, try to re-route the path. This causes inefficiency in factories and makes AMR unsuitable. AMRs are limited to Fully automated facilities (factories or warehouses) with minimal human involvement. Scaling the deployment of AGVs and AMRs becomes problematic due to the need for extensive customization and reconfiguration as the operational environment expands or evolves.

[0010] Therefore, there is a need in the art with a hybrid self-driving ground vehicle and a method thereof to solve the above-mentioned limitations.SUMMARY OF THE INVENTION

[0011] An aspect of the present invention is to address at least the above-mentioned problems and / or disadvantages and to provide at least the advantages described below.

[0012] Accordingly, in one aspect of the present invention relates to hybrid selfdriving ground vehicle (100) for autonomous navigation. The hybrid self-driving ground vehicle (100) comprises a mobility platform (102) configured to actuate movement of the vehicle, an odometry sensor system (104) configured to continuously estimate translational and rotational motion of the mobility platform, a visual sensing system (106) comprising plurality of two-dimensional camera configured to acquire image data of an environment, an obstacle detection system (108) configured to detect dynamic and static obstacles, one or more processors (110, 112) operatively coupled to the odometry sensor system, the visual sensing system, and the obstacle detection system, wherein the one or more processorsconfigured to: select between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode, continuously localize the vehicle using the odometry sensor system, intermittently perform environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment and autonomously navigate the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

[0013] Another aspect of the present invention relates to a method (900) for autonomous navigation of a hybrid self-driving ground vehicle. The method (900) comprises actuating (910) movement of the vehicle via a mobility platform, continuously (920) estimating translational and rotational motion of the mobility platform using an odometry sensor system, acquiring (930) image data of an environment with a visual sensing system comprising a plurality of two-dimensional cameras, detecting (940) dynamic and static obstacles using an obstacle detection system, selecting (950) between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode, continuously (960) localizing the vehicle using the odometry sensor system, intermittently (970) performing environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment, and autonomously (980) navigating the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

[0014] Other aspects, advantages, and salient features of the invention will become apparent to those skilled in the art from the following detailed description, which, taken in conjunction with the annexed drawings, discloses exemplary embodiments of the invention.BRIEF DESCRIPTION OF ACCOMPANYING DRAWINGS

[0015] The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference features and modules.

[0016] Figure 1 illustrates a system block diagram of Hybrid Self-driving Ground Vehicle (HSGV) according to an exemplary implementation of the present invention.

[0017] Figure 2 illustrates a flow diagram of AGV Training Mode of HSGV according to an exemplary implementation of the present invention.

[0018] Figure 3 illustrates a flow diagram of AGV Navigation Mode of HSGV according to an exemplary implementation of the present invention.

[0019] Figure 4 illustrates a flow diagram of AMR Training Mode of HSGV according to an exemplary implementation of the present invention.

[0020] Figure 5 illustrates a flow diagram of AMR Navigation Mode of HSGV according to an exemplary implementation of the present invention.

[0021] Figure 6 illustrates a depth function output of an image using the Depth Al models according to an exemplary implementation of the present invention.

[0022] Figure 7 illustrates a flow diagram of object training according to an exemplary implementation of the present invention.

[0023] Figure 8 illustrates a flow diagram of the concept of Keyframe-based Intermittent SLAM according to an exemplary implementation of the present invention.

[0024] Figure 9 illustrates a method for autonomous navigation of a hybrid selfdriving ground vehicle according to an exemplary implementation of the present invention.

[0025] It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative methods embodying the principles of the present disclosure. Similarly, it will be appreciated that any flow charts, flow diagrams, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.DETAILED DESCRIPTION OF THE INVENTION

[0026] The various embodiments of the present invention / disclosure describe techniques in the hybrid self-driving ground vehicle, aiming for widespread adoption across various industries for material transportation indoors and outdoors within a campus.

[0027] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of exemplary embodiments of the invention as defined by the claims and their equivalents. It includes various specific details to assist in that understanding but these are to be regarded as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. In addition, descriptions of well-known functions and constructions are omitted for clarity and conciseness.

[0028] The terms and words used in the following description and claims are not limited to the bibliographical meanings but are merely used by the inventor to enable a clear and consistent understanding of the invention. Accordingly, it should be apparent to those skilled in the art that the following description of exemplary embodiments of the present invention are provided for illustration purpose only and not for the purpose of limiting the invention as defined by the appended claims and their equivalents.

[0029] It is to be understood that the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to “a component surface” includes reference to one or more of such surfaces. References in the specification to “one embodiment” or “an embodiment” mean that a particular feature, structure, characteristic, or function described in connection with the embodiment is included in at least one embodiment of the invention. The appearances of the phrase “in one embodiment” in various places in the specification are not necessarily all referring to the same embodiment.

[0030] By the term “substantially” it is meant that the recited characteristic, parameter, or value need not be achieved exactly, but that deviations or variations, including for example, tolerances, measurement error, measurement accuracy limitations and otherfactors known to those of skill in the art, may occur in amounts that do not preclude the effect the characteristic was intended to provide.

[0031] Figures discussed below, and the various embodiments used to describe the principles of the present disclosure in this patent document are by way of illustration only and should not be construed in any way that would limit the scope of the disclosure. Those skilled in the art will understand that the principles of the present disclosure may be implemented in any suitably arranged communications system. The terms used to describe various embodiments are exemplary. It should be understood that these are provided to merely aid the understanding of the description, and that their use and definitions in no way limit the scope of the invention. Terms first, second, and the like are used to differentiate between objects having the same terminology and are in no way intended to represent a chronological order, unless where explicitly stated otherwise. A set is defined as anon-empty set including at least one element.

[0032] In the following description, for purpose of explanation, specific details are set forth in order to provide an understanding of the present disclosure. It will be apparent, however, to one skilled in the art that the present disclosure may be practiced without these details. One skilled in the art will recognize that embodiments of the present disclosure, some of which are described below, may be incorporated into a number of systems. However, the systems and methods are not limited to the specific embodiments described herein. Further, structures and devices shown in the figures are illustrative of exemplary embodiments of the presently disclosure and are meant to avoid obscuring of the presently disclosure.

[0033] Various embodiments of the present invention are further described with reference to FIG. 1 to FIG. 9.

[0034] The present invention discloses that autonomous vehicle / hybrid self-driving ground vehicle uses Keyframe-based Intermittent SLAM based on 2D camera and simplified 2D mapping, enabling higher energy saving for longer battery life and making it easier for users to understand and interact with the system.

[0035] In the present invention, HSGV denotes Hybrid Self-driving Ground Vehicle which is the proposed autonomous vehicle system.

[0036] The primary objective of the present invention is designed to address the existing gaps and limitations in the field of autonomous vehicles for material transportation, aiming for widespread adoption across various industries. Therefore, the key goal of the present invention is to enhance automation for operational efficiency, to increase scalability and flexibility to a certain extent, to increase reliability, quality and safety while being economically feasible in nature.

[0037] According to the present invention, the proposed autonomous vehicle system aims to revolutionize material transportation within facilities and campuses by enabling fully autonomous operation minimizing human intervention. This advanced system ensures streamlined operations, prevents manual errors, and significantly enhances overall efficiency. Designed for seamless integration, these vehicles can be incorporated into existing infrastructures and workflows with minimal modifications, allowing for swift adoption. The vehicles are adaptable to various environments, from tightly controlled indoor settings to more variable outdoor conditions, making them versatile and suitable for diverse operational contexts. Additionally, the scalable nature of the solution ensures that it can be expanded easily and cost-effectively, accommodating both small and large-scale operations without incurring significant additional costs.

[0038] According to the present invention, the system disclosed herein has increased reliability, quality, and safety as key features. Advanced diagnostics and predictive maintenance capabilities foresee and prevent potential failures, reducing downtime and extending equipment lifespan. Robust navigation systems ensure precise and error-free material transport, while dynamic obstacle avoidance enhances safety through intelligent detection and decision-making. The HSGV handle materials consistently, maintaining or enhancing product quality and reducing the risk of damage during transport. By utilizing data collected from the autonomous vehicles, the system can identify bottlenecks and optimize workflow efficiency, leading to overall process improvement. Economically, the solution offers a cost-effective implementation with a clear and rapid return on investment, appealing to industries of various sizes and budgets. Long-term cost savings are realized through reduced labor costs, lower maintenance expenses, and minimized material wastage, making this HSGV system an attractive and sustainable option for modernizing material transport operations.

[0039] The present invention discloses hybrid self-driving ground vehicle and a method for autonomous navigation. The vehicle is configured to selectively operate in an Automated Guided Vehicle (AGV) mode or an Autonomous Mobile Robot (AMR) mode using a common hardware and software architecture. The system employs one or more monocular two-dimensional cameras and an odometry sensor system to achieve continuous localization, while environment mapping is performed intermittently at selected keyframes rather than continuously. A navigation-oriented representation of the environment is generated without continuous three-dimensional reconstruction, enabling accurate navigation with reduced computational load, energy consumption, and data bandwidth. In AGV mode, navigation is performed along pre-trained paths without default autonomous rerouting, while obstacles are detected and alerts are generated for human intervention. In AMR mode, automatic exploration, object-based localization, and controlled re-routing are enabled. Positional accuracy is maintained through intermittent calibration using markers, known static objects, or virtual reference points. The disclosed system enables reliable autonomous operation in human-shared environments without reliance on three-dimensional LiDAR or stereoscopic vision systems.

[0040] Figure 1 illustrates the system block diagram of Hybrid Self-driving Ground Vehicle (HSGV) according to an exemplary implementation of the present invention.

[0041] The figure illustrates the system block diagram of HSGV (100). According to the present invention, the proposed invention stands as an advanced autonomous vehicle that integrates the functionalities of Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs). The primary concept is to simplify autonomous navigation through Keyframe-based Intermittent SLAM using cost-effective and efficient 2D cameras as the primary navigation sensor, making it highly power efficient, easily deployable and operable in various industrial environments. The present invention vehicle operates under a dualmode system, i.e., AGV mode and AMR mode, allowing it to adapt its navigation strategy based on the operational requirements.

[0042] In one embodiment of the present disclosure / invention, HSGV (100) architecture comprises: mobility platform - motor, driver and encoder system (102), plurality of sensor system (104, 108, 114, 122, 124), camera system (visual sensing system) (106),one or more processor (110, 112) light and sound system (120), one or more external subsystem (116), GPS and IMU (118) and touchscreen display (126).

[0043] The hybrid self-driving ground vehicle (100) for autonomous navigation comprises a mobility platform (102) configured to actuate movement of the vehicle, an odometry sensor system (104) configured to continuously estimate translational and rotational motion of the mobility platform, a visual sensing system (106) comprising plurality of two-dimensional camera configured to acquire image data of an environment, an obstacle detection system (108) configured to detect dynamic and static obstacles, one or more processors (110, 112) operatively coupled to the odometry sensor system, the visual sensing system, and the obstacle detection system, wherein the one or more processors configured to: select between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode, continuously localize the vehicle using the odometry sensor system, intermittently perform environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment without continuous three-dimensional reconstruction which is sufficient for navigation accuracy, and continuous re-computation of a global environment map is avoided and autonomously navigate the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

[0044] In an embodiment, the navigation-oriented representation comprises representation of static obstacles, traversable free space, semantic destinations, and reference relationships, without storing a dense geometric map. The navigation-oriented representation enables reduced computational load, energy consumption, and data bandwidth relative to continuous SLAM-based systems.

[0045] The vehicle, in the AGV mode, one or more processors are configured to: compute a navigation plan along a user-trained path defined by a plurality of stop points and vector relationships between the stop points, detect and avoid collision with the obstacles detected by the obstacle detection sensor system and realign position and orientation of the vehicle at the stop points using at least one of a physical marker, a pre-trained static object, or a virtual reference point upon user confirmation, and wherein the realignment resetsaccumulated positional and orientation error, wherein object-based localization is used as a fallback localization mechanism independent of physical markers.

[0046] In another embodiment, in the AGV mode, the vehicle follows a pre-trained fixed path without any real path marking on the floor (like painted lines, tapes, magnetic tapes etc.,) and comes to a standstill upon detecting an obstacle in the pre-trained path and generate alerts through at least one of visual indicators, audible signals, or notifications to a supervisory system, without performing autonomous re-routing.

[0047] In AGV navigation mode, the vehicle follows trained paths, detects obstacles and halts safely, communicates issues for human resolution, maintains accuracy through reference alignment, enables re-routing only via explicit mode transition. This design delivers the predictability of traditional AGVs combined with the flexibility of autonomous systems, without inheriting the weaknesses of either approach.

[0048] The vehicle, in the AMR mode, one or more processors are configured to: compute a navigation path on a self-trained path or explored traversable free space to a user-specified semantic destination over the two-dimensional navigation-oriented representation, detect and avoid obstacles detected by the obstacle detection sensor system and reroute around detected obstacles with intermittent calibration without performing full environment remapping.

[0049] In another embodiment, in the AMR mode, the processor is configured to compute a bypass trajectory around a detected obstacle based on at least one of pre-trained object dimensions or newly detected obstacle geometry, and merge with a previously computed path.

[0050] In AMR navigation mode, the vehicle autonomously navigates within a previously trained environment using continuous awareness-based localization combined with intermittent calibration, rather than continuous simultaneous localization and mapping (SLAM). This embodiment enables high-accuracy autonomous navigation without relying on traditional continuous SLAM, three-dimensional LiDAR, or continuous three-dimensional environmental reconstruction.

[0051] In the present invention, the selected keyframes comprise at least one of home position, navigation initiation, arrival at a user defined stop point, arrival at a virtual reference point, detection and alignment with a fiducial marker, recognition of a pre-trained static object, obstacle-induced stopping, rerouting events, or completion of navigation.

[0052] The one or more processors (110, 112) comprises at least one primary processor and one or more secondary processors. In the present invention, one or more processors (110, 112) comprise a primary processor (110) executing navigation mode management, task planning, user interface handling, and keyframe mapping control, and at least one secondary processor (112) executing mobility control, safety monitoring, sensor preprocessing, and external subsystem interfacing.

[0053] Primary processor (110) is responsible for: 1) Running Navigation software, 2) Sending and receiving signals and data from the Secondary processor, 3) Sending commands to the motor driver system in order to move, change direction, stop the motors connected to it and processing data related to status / configurations of motors and encoder feedback, 4) Processing manual inputs from touchscreen panel and displaying the user interface which may include the current status, the next target, list of inputs, etc. 5) Processing image and / or video data coming from the camera system, 6) Processing data related to obstacles / objects such as their presence or absence, their distance from the vehicle, their shape or size, etc. from the obstacle detection system, 7) Processing signals related to manual override, collision, accidental fall or anything related to safe operation from the safety sensor system, 8) Processing odometry sensor system data and corelate with other positional data, 9) Sending commands to external sub-system to trigger start / stop / pause one or more actions and processing data related to feedback from external subsystem.

[0054] In an embodiment, the primary processing unit (110) is configured to perform high-level decision-making, navigation logic, task execution, user interface processing, and coordination of system-level behaviors. The primary processing unit executes software modules including, but not limited to: navigation mode management (AGV mode, AMR mode), task planning and sequencing, keyframe-based intermittent mapping control, user interaction handling, communication with external systems and fleet management services.

[0055] Secondary processor (112) is responsible for: 1) Sending and receiving signals and data from the primary processor, 2) Processing signal / data from the Payloadsensor system and calculates the total weight of the items placed on top of the vehicle, 3) Processing data related to obstacles / objects such as their presence or absence, their distance from the Vehicle, their shape or size, etc. from the obstacle detection system, 4) Processing signals related to manual override, collision, accidental fall or anything related to safe operation from the safety sensor system, 5) Sending commands to external sub-system to trigger start / stop / pause one or more actions and processing data related to feedback from external subsystem, 6) Sending signals to turn ON / OFF / BLINK light system and turn ON / OFF / change the sound system to signify the different statuses of Vehicle operation like different types of movement (forward, backward, right turn, left turn), Battery low alert, obstacle alert, other alerts etc. 7) Processing signals such as battery percentage, charging status, from battery system and sending the data to the touchscreen display through the primary processor, 8) Processing image or / and video data coming from the camera system (applicable in case of more than one secondary processor).

[0056] In an embodiment, the secondary processing units (112) are configured to perform time-critical, sensor-facing, and safety-related functions, including: acquisition and preprocessing of sensor data, motor control and actuator command execution, safety monitoring and emergency handling, external subsystem interfacing, signal conditioning and real-time control loops. This distributed processing architecture reduces computational load on the primary processor, improves fault tolerance, and enables deterministic real-time response for safety-critical operations.

[0057] In an embodiment, the one or more processors are further configured to perform object-based localization by using one or more pre-trained static objects as spatial reference anchors to determine or correct the position and orientation of the vehicle independent of physical markers or continuous environment mapping.

[0058] In figure 2, Object recognition thread, monocular depth thread, estimation of distance and orientation of static objects, creation of object points on map imply the Object based mapping process. In Figure 3, trilateration-based localisation refers to object-based localisation process.

[0059] In one embodiment, the hybrid self-driving ground vehicle is configured to perform object-based localization, wherein the vehicle determines or refines its position within an environment by recognizing and spatially relating itself to pre-trained staticobjects, without requiring continuous simultaneous localization and mapping or dense geometric reconstruction. This embodiment enables accurate localization in environments where: physical markers are sparse or unavailable, traditional SLAM is undesirable or impractical, human activity causes frequent short-term visual changes.

[0060] According to the present invention, a reference object is a static, spatially stable object whose position is known or learned during a training phase and stored in the system database. Reference objects may include, but are not limited to: machines, racks or shelves, pillars or columns, cabinets, workstations, fixed infrastructure elements. Objects that are likely to move or change position are intentionally excluded from reference classification.

[0061] The plurality of sensor system comprises at least one battery sensor system (122), at least one payload sensor system (124), at least one odometry sensor system (104), one or more sets each of Safety sensor system (114) and obstacle detection system (108) in the entire architecture to provide multi levels of safety and minimize chances of positional error. Each safety sensor system (114) consists of limit switches, emergency stop switches, vibration sensors, gyroscope, magnetometer, accelerometer, etc. or a combination of these. The safety sensor system further includes emergency stop switches, collision sensors, tilt or vibration sensors, and inertial safety triggers, enabling immediate transition to a safe state upon detection of abnormal or unsafe conditions or loss of localization. Each obstacle detection system (108) may include one or more of time-of-flight sensors, ultrasonic sensors, infrared sensors, two-dimensional LiDAR sensors or combinations thereof, and is configured to provide short-range and mid-range proximity awareness independent of lighting conditions. Payload sensor system (124) consists of one or more load cells depending on the rated load carrying capacity of vehicle. Its purpose is to sense the total load and give the information to the Secondary processor. This helps in preventing overloading of the motors. Battery Sensor System (122) consists of two parts: One is Battery percentage sensing and the other is Charging detection. These information are used by the secondary processor to determine when to send battery low alerts to the primary processor, to predict whether task at hand can be completed with the current battery capacity, detect any anomaly in charging dock, if its charging then to make sure, no more tasks are taken until charging is completed etc. Odometry sensor system (104) may include wheel encoders, inertial measurement units (IMUs), other odometry sensors, or combinations thereof, and isconfigured to continuously estimate translational and rotational motion / displacement of the vehicle. The odometry sensor system detects the relative movement of the vehicle with respect to the previous position. It not only measures the distance but also the angle and provide a feedback to the system for accurate path following during training and navigation operations.

[0062] Motor, Driver and Encoder System comprises: two or more sets of driver connected to motor, encoder and wheel. The Motor, Driver and Encoder System collectively referred as mobility platform (102) configured to actuate movement of the vehicle. The present disclosure / invention system can have only even no. of such sets. Also, in some cases, there can be belt driven or chain driven tracks instead of wheels for usage in outdoors and on uneven surfaces. And in some other cases, for omni directional movement or better directional flexibility, there can be multiple motors and corresponding drivers and encoders for a single wheel. Encoder is used for feedback and can be integrated with the motor or externally mounted based on the type of motor used.

[0063] Camera System or visual sensing system (106) consists of plurality of monocular 2D cameras, in one embodiment, the plurality of monocular 2D cameras comprises at least five cameras: Front, Back, Left side, Right side and Down facing Cameras. Depending on the application, few cameras can be deactivated while deployment, for saving power and bandwidth. Further, in the present invention, the camera activation is selectively controlled based on user application, type of environment, the operational mode (AGV or AMR) and selected keyframe requirements. In an embodiment, only 2 or more cameras may be present in the system. However, the accuracy improves with the presence of more cameras. In another embodiment, the camera activation and processing frequency are manually or adaptively / dynamically adjusted based on operational mode, environmental complexity, or task criticality.

[0064] Light and sound system consists of multi colour LEDs, each colour or blinking pattern signifying a certain state of vehicle and speaker / alarm signifying alerts.

[0065] One or more External subsystem (116) consists of additional electronics or / and mechanical system which form part of HSGV for performing a certain application. For example, conveyor system, pallet carrier, lifting mechanisms, robotic arms, pallet handling systems, inspection or sensing modules upon reaching designated stop points.These external sub systems get triggered based on the signals generated by the vehicle upon reaching a certain destination. Also, upon completion of a certain task by these subsystems, they send signals to the primary or secondary processor so that the next decision can be taken.

[0066] GPS and IMU consist of GPS modules which are used for global positioning of vehicles. IMU are inertial measurement units and include accelerometers, gyroscopes, magnetometer etc. These are not mandatory for autonomous mobility but prove useful in providing added features of vehicle.

[0067] Figure 2 illustrates the flow diagram of AGV Training Mode of HSGV according to an exemplary implementation of the present invention.

[0068] The figure illustrates the flow diagram of Automated Guided Vehicles (AGV) Training Mode of HSGV. HSGV can be configured to operate in either Automated Guided Vehicles (AGV) or Autonomous Mobile Robots (AMR) mode through the User Interface (UI) application running on the vehicle or any other computing device connected to the same network. If configured as AGV, the vehicle needs to be manually trained first before one start using it. In AGV mode, navigation path of the vehicle is manually trained using a handheld remote, teach -pendant-like interface, or user interface. Manual training refers to creating a 2D map of the floor of a facility and a virtual path on it for the movement of the vehicle. Virtual path is a line between stop points or virtual points on the map. These virtual / stop points are the positions where the HSGV needs to pause / stop.

[0069] In this process, initially, the floor layout is uploaded onto the UI or an approximate floor layout is created on the UI using basic parameters like shape, length and width of the space. Then the home position is marked on the layout. Home position is the position on the floor where the HSGV is available whenever it is idle. It is generally the first point on the floor where the training starts. Home can also be a charging station where the vehicle can come for charging when the battery is low. Origin of the map is home position by default, which the user can change.

[0070] It is recommended to place the HSGV at home position before the start of the training. For every stop point, an identifier is must. The identifier can be a special 2D marker printed on a flat surface such as paper, plastic, metal and can be pasted on the floor, on thewall, on objects, on racks. The 2D marker can also be printed on a card (like ID card) and can be hung on a nail, from the ceiling or from racks. These markers are special as they help in not only identifying the current position, but also distance and orientation of the vehicle / system with respect to the marker from different camera angles. The markers, each with its unique Id, can be generated using the UI. Generally, one marker is used to define one stop point. If the user wants the vehicle to stop at a point and there is no possibility of pasting an identifier, then it is called virtual point. If the vehicle detects a pre-trained static object in the view, then the object position is also called virtual point only if user confirm that there is a path towards the object.

[0071] In an embodiment, the markers are used only at selected semantic locations or keyframes and not in a uniform grid or markers are not necessarily used at uniform distance on a line.

[0072] Once the stop points are defined by pasting marker physically. Then training button (physical button on the vehicle or handheld remote, or software button on User interface (UI)) is pressed to mark the change of mode to training. Instructions are shared with the user through texts on the UI or / and through sound and light signal. User press buttons on the remote as per instruction to move the vehicle forward, backward, right, or left. When the vehicle is in motion, this motion is recorded through the feedback from Motor and Encoder system. Once the motion stops, user presses a certain button to indicate the arrival of a stop point. User gets an option on the UI to choose whether to look for a marker or not. If user selects yes, then marker recognition function is activated disabling the remote user access.

[0073] Marker recognition function is a modified geometric transformations and marker-specific image processing techniques. Vehicle analyses the input video stream from the camera and once the marker is detected and recognized, then marker alignment function is activated. If a marker is not detected, then remote access is enabled again and user is instructed to press buttons to move the vehicle such that marker is in the view of the HSGV. The marker can be seen on the camera view on the UI as well so that user knows that he / she is doing it right.

[0074] Marker alignment function is an Al model developed for vehicle to move such that the center of the image frame coincides with the center of the camera frame at acertain orientation with respect to x axis of the marker. At 0 °, x-axis of image frame is parallel to the x-axis of the marker. The orientation angle for alignment is generally decided automatically based on the angles at which markers are pasted on the workfloor and also the direction of motion from the previous marker position. This orientation angle matching is done to nullify any position errors that the vehicle might accumulate between two markers or between two virtual / stop points. Errors may occurs due to uneven surface on workfloors, slippage due to oil / water spills, deviation due to small objects lying on the floors (mimicing uneven surface), anything which may cause the wheels to change direction.

[0075] After Marker is detected and aligned, the current position of HSGV is calculated with respect to the origin. Then it’s position vector and the camera name (which is used for capturing marker information) are saved. This position is marked as stop point and is automatically labelled with an unique name which can be edited and saved by the User as per relevance. In AGV mode, vehicle can be deployed with only one camera or more cameras depending on the application: down camera is to detect floor markers, front camera is to detect the environment, obstacles and markers on objects, side cameras are to detect environment and obstacles while turning and also to detect markers on the wall, racks, objects, back camera is used like front camera during backward motion.

[0076] If only bottom camera is active, then after marker alignment, the vehicle is ready for the next motion (path). If front camera, back or side cameras or all of them are active, then the process is longer as user has more options to choose from. In some cases, multiple markers may be captured if markers are nearby on a wall or on racks. If two or more markers are detected from the same camera, by default, marker which is closest will be taken into consideration for alignment. If two or more cameras detect markers at the same time, user gets to choose which camera frame to be used for marker alignment. If it is left camera, the vehicle will perform additional alignment process like aligning parallel to the marker at a pre-defined distance from the left wall or rack or object where the marker is pasted. Similar would be the case with right, front or back camera. But if it is down camera, the distance of the camera from the floor is fixed therefore, the vehicle only performs basic alignment like bringing the vehicle centre to the centre of the marker at the predefined angle with respect to marker x-axis. To avoid multiple marker detection, uniform marker pasting is recommended such as only floor markers or only wall markers or only object markers etc.In these cases, only down camera can be active, only side cameras can be active and only front or back cameras can be active, respectively during training.

[0077] During training, if the vehicle motion stops at a virtual point, a point where there are no physical markers, user indicates that by pressing a certain button. Then instead of marker recognition function, the vehicle cameras capture images of its environment, extract their features using a hybrid model involving SIFT and ORB and save them. Then object recognition function is activated.

[0078] If a pre-trained static object is detected, then vehicle processes the image using depth estimation function and calculates the position of the object with respect to the origin. After this, user gets a choice to track the object. If the no. of objects are more than one, each of their positions are calculated and then users gets an option on the UI to select one of them for tracking. If tracking is not selected, then only the position vector of the current position is saved, automatically labelled and marked as virtual point. Position vectors of all the identified objects are also saved, auto labeled with unique names and object positions are saved for reference during autonomous navigation. If no object is detected, the vehicle directly saves the current position as virtual point with the position vector and automatic labelling.

[0079] If tracking is selected, then user gets to choose which obj ect to track. Tracking is generally done for the purpose of defining the path of the current virtual point to the object automatically using the calculated shortest path. HSGV moves on the estimated path towards the object and stops at an accurate pre-set distance from the nearest surface ofthe object. For example, if the detected object is a machine or a manufacturing unit, HSGV can track and accurately arrive one feet away from it. Now, if there is a marker pasted on the object, HSGV aligns with the marker and saves the current position as stop point. Otherwise, it releases control to the user so that user can take the vehicle to the next stop / virtual point.

[0080] The positional accuracy of the vehicle at a virtual point depends on it accuracy in terms of position and orientation at the previous stop point and also the no. of reference pre-trained static objects in the camera views. The accuracy can reduce if the environment around the virtual point is plain (no pre-trained static objects), prone to change frequently or to get crowded with people such that visibility is hampered. Therefore, virtual points are used only if marker is not an option. Although virtual points has limitations asmentioned above, it is an intelligent feature which does solve the problem of navigating a space without creating any physical markers.

[0081] After every point to point training, the position vectors of virtual / stop points and objects are plotted on the map which was selected in the beginning of the training. Then the virtual / stop points are connected by lines on the map, to mark the path which was trained. If a path between two virtual / stop points is not trained, then line does not form on the map. Upon the completion of the training, a 2D map of the workfloor is created and saved with an user preferred name. User indicates the completion of training by pressing a certain button. User also gets an option to delete or edit the trained data. If the trained data is deleted, path or points will be deleted from the map as well.

[0082] The map is based on a coordinate system where all the position vectors refer to the origin (0,0). User can add more stop points or virtual points on an existing path by retraining the vehicle or by editing the map on the UI. User needs to open an existing map, place the pointer at the path position and click to add new stop points, add marker id, label and save. While doing this, the coordinates of the point will be made visible on the UI so that user can take the right decision. This marker needs to be pasted on the same coordinates on the actual floor. To add a new virtual point, HSGV should be brought to a known position first. HSGV does not retrains a virtual / stop point which is already saved. But HSGV can be put on training mode again by selecting an existing trained floor map. In this case, a new virtual / stop point can be created by physically moving HSGV on an existing path but stopping before the next virtual / stop point as per map.

[0083] Another very useful feature is that the entire training can be done without physically moving the vehicle on the work -floor. This is possible by uploading training data which relates markers with stop point names and camera name, and relates any two virtual / stop points with the distance vector which is distance with orientation information. Optionally, the training data can be created on UI directly. This also provides the user a possibility to train the vehicle on the same floor again on a different path or extending the path from an existing virtual / stop point.

[0084] In an embodiment, user defined motor control on desired path implies teach pendant type manual training. Use of markers implies accuracy control through alignment. Creation of stop points by saving the position vectors implies the relationship between them.If markers are not available, virtual points are created to define the keyframes. To improve localisation accuracy at virtual points, object-based localisation is used. So, HSGV is following a fixed path although there is no real line or tape on the floor. Markers are used only to improve accuracy. In an embodiment, markers can be omitted and object recognition can be used instead.

[0085] Figure 3 illustrates the flow diagram of AGV Navigation Mode of the hybrid self-driving ground vehicle according to an exemplary implementation of the present invention.

[0086] The figure shows the flow diagram of Automated Guided Vehicles (AGV) Navigation Mode of the hybrid self-driving ground vehicle. If the vehicle is configured as AGV, and at least one training data exists for at least one map (floor layout) in the system, the user can switch to Navigation mode . In this mode, user can operate the vehicle to navigate from one virtual / stop point to another automatically. The user can input a destination on the UI, select the floor map where the vehicle is located and trigger the HSGV to move from its current position to the given destination. If no training data is found for, user is instructed to go to training mode or select another map. This check allows the user to be sure if the selected map is correct and also enables user to train HSGV on the existing map / workfloor if no previous training has happened.

[0087] After receiving the destination, HSGV first identifies its current position. It retrieves the last known position from its own memory and the local server. If both match, means HSGV was not moved from its last known position. If both do not match or if data from server is not available, HSGV last position is unknown. Alert is raised through light, sound or / and notification on the UI. User gets the control to move the vehicle to a stop / virtual point manually and trigger a certain button when done. If user takes HSGV to a stop point, then marker recognition function is activated, to identify its current position. If marker is detected, then marker alignment function is activated. Default alignment is when there is no angle between the x axis of the camera frame and the marker frame. The current position of HSGV in relation to the marker is updated on the map. Also, it labels its current position temporarily as source.

[0088] If Marker is not detected, marker search function is activated where HSGV moves in its vicinity and tries to detect the marker. If marker is not detected, HSGV activatesmarker search function again and increments a counter. If the count goes beyond a certain predefined value, HSGV raises an alert in the form of light, sound signal or / and UI notification. User gets the control to move the vehicle to a stop / virtual point manually and trigger a certain button when done. The process continues as mentioned in Paragraph

[0087] .

[0089] If user takes HSGV to a virtual point, then image matching function is activated. HSGV captures the images of its surrounding through all the active cameras and extracts their features and tries to match with trained database. If the features match upto a certain predefined threshold, then its is considered a match and HSGV is able to identify its current position and updates the same on the map and labels it temporarily as source. If image matching fails, then Object recognition function is activated. HSGV tries to detect and recognize minimum 3 pre-trained static objects in the input stream of its cameras. Once, at least 3 objects are recognized, Depth function is activated through which their shortest distances from HSGV are calculated. Since, the position of the recognized objects are known from training database, the current position of HSGV is calculated through trilateration. HSGV then updates its current position on the map and labels it temporarily as source.

[0090] It is very important that each of the objects on the workfloor are unique. Else, this method may result in multiple possible positions of HSGV and therefore, user control will be activated, after raising alerts. This enables the user to position the HSGV to the nearest known position and trigger a certain button when done. The process continues as mentioned in Paragraph

[0087] and followed thereon. If 3 objects are not detected, object search function is activated where HSGV moves in its vicinity and runs object recognition function in parallel. If 3 objects are not detected, HSGV again goes to object search function and increments a counter. If the count goes beyond a certain predefined no. x, HSGV raises alerts and user gets the control to move HSGV to a known position and trigger a certain button when done. The process continues as mentioned in Paragraph

[0087] and followed thereon.

[0091] Once source is found, HSGV checks if the source is same as user input destination. If it is true, then HSGV has successfully arrived at the destination and waits for another user input for navigation mission. Else, it computes the shortest path from the source to the given destination via trained virtual / stop points. If any two virtual / stop points are notconnected by a trained path, that path is not considered for shortest path calculation. Algorithm like A star is used for this process. Once shortest path is calculated, Vehicle makes the next via virtual / stop point as the next destination and starts moving towards it, while updating its odometry data continuously on the map. For example, the shortest path may be 1 to 2 to 5 to 10, where 1,2,5,10 may be stop or virtual points and the path between them is retrieved from the training database. Here, 1 is the source, 2 the next destination. Upon reaching 2, 2 becomes the source, 5 as the next destination and same way it continues until the source and destination are the same. If the distance vector between 1 and 2 is 10 meters, angle 0 degree, it means HSGV will move forward until the estimated distance (lOmetres) minus travelled distance (odometry data) is zero.

[0092] During this motion if Vehicle detects an obstacle within a certain distance, it stops moving and raises alert in the form of light, sound signals, or / and UI notification. Vehicle is intelligent in determining what type of sound and UI notification to activate based on the recognized obstacle. If the obstacle is a human being, Vehicle activates human like sounds, where it can communicate in human voice. For example, Vehicle can say “Please give me way.” If Human responds saying “I need 1 minute”, Vehicle processes the data through the onboard sound sensors and predicts that there will be 1 minute delay in completing the current task. The complexity of the communication and corresponding prediction and decision making depends on the NUP training models and Generative training models of HSGV.

[0093] If the obstacle is any other object or human being, HSGV waits for the obstacle to move only until a predefined wait time. If the obstacle continues to exist after the wait time, then another alert is send to user to either take control of HSGV or move the obstacle from the path. Once the obstacle is no longer in the vicinity of HSGV, it resumes its navigation task. Upon reaching the destination, it triggers alerts regarding completion of the task and that the Vehicle is ready for another navigation task and process continues as mentioned in paragraph

[0086] and followed thereon.

[0094] Although, there is one real destination, using the via virtual / stop points as temporary destinations prevents Vehicle from accumulating positional errors over long distances and while taking turns. This is possible because at every temporary destination vehicle aligns with the target identifier (marker, image, objects) accurately. This ensuresvehicle provides highest accuracy throughout its navigation journey. The best thing is that there is no requirement to create virtual / stop points frequently on a straight path, but mainly at junctions or turning points. Therefore, even in a large facility, the total no. of markers used could be very small. For example, in a 10000 sqft space if there are two main stop points for Vehicle where packages are picked up and dropped, then the total no of markers used could be just 2 if there is a straight path between the two stop points. The no. of markers can increase to 3,4 or 5 if the path involves 1, 2 or 3 turns respectively.

[0095] AGV navigation mode is very elaborate and gives lots of options to the users to utilize the HSGV to carry out various tasks. Through the Task manager page, user can create, edit or delete tasks. Creating tasks involve several parameters like setting the source, destination, add in between stops, wait time, run now or schedule later, similar to booking cabs through apps or generally how people navigate using GPS based maps. For example: A task T1 is created with Source as A, Destination as C, added stop as B and wait time of 5sec and repeat as 10. When T1 task is run immediately, HSGV vehicle first finds its current position, for example E, updates it as source. It takes A as the destination and starts moving towards A. After reaching A, HSGV updates its source as A, and destination as B, waits for a maximum of 5sec so that some action like manual loading and unloading can complete. If the action completes before 5sec, then user presses a trigger button so that Vehicle can start moving towards B. And the same steps follow until the vehicle reached C after completing 10 rounds of motion from A to C via B and receives an action completion trigger or 5sec whichever is shorter. There is also a provision to add automatic actions at every source and destination to enable automatic loading and unloading of packages on HSGV. This is done by automatically sending a trigger signal to an external subsystem like a robotic arm, conveyor, fork-lift etc upon reaching the source or destination. This gives the flexibility to use Vehicle not only for transportation but also automatic pickup and delivery, and many other functions like Automatic quality inspection, Automatic inventory audit etc.

[0096] In an embodiment, during navigation in AGV mode, the system performs shortest path plan from among the available trained paths. To do that, first it determines its current location through a sequence of steps which involves marker-based localisation and if failed then, object-based localisation. When obstacle is detected, it waits until it is removed. This implies that there is no re-routing in this mode. Also, alert is raised which implies manual over-ride can be performed although it is not explicitly shown in the flowdiagram. In manual over-ride, mode can be changed to AMR mode where re-routing can be performed if the situation demands.

[0097] In an embodiment, the selection between the Automated Guided Vehicle mode and the Autonomous Mobile Robot mode is adjusted dynamically based on task requirements and environmental conditions.

[0098] Figure 4 illustrates the flow diagram of Autonomous Mobile Robots (AMR) Training Mode of the hybrid self-driving ground vehicle according to an exemplary implementation of the present invention.

[0099] The figure illustrates the flow diagram of Autonomous Mobile Robots (AMR) Training Mode of HSGV. In AMR mode, training of Vehicle is automatic. Training is like mapping an area or layout such that wherever it goes, Vehicle can localize itself accurately with respect to a reference (origin). In this mode, first a floor layout or an outdoor campus layout is uploaded in a particular format in the vehicle system. This defines the outer boundary of the area to be mapped. Then the Vehicle is positioned within this area such that its approximate position and orientation with the origin and a virtual x axis through origin are known. This is also called home position and chosen in such as way that HSGV is at a comer of a workfloor or maximum 5 meters away from a wall. Origin and Home position with respect to origin is first manually marked on the layout by the user. After manual marking is complete, HSGV automatically calculate the distance between Vehicle and the nearest wall and updates its home position coordinates with respect to the layout.

[0100] Vehicle then starts the training by capturing the images from all the active cameras in different directions. It extracts the features of these images by using ORB / SIFT algorithms. The extracted features are compared with the feature database and if there is a match beyond a certain threshold, it is considered as duplicate. If duplicate features are detected, then it checks whether any of the traversable free spaces connected to this feature saved in the map database, is not traversed. If all are traversed, then training is deemed complete. If there is even a single traversable path available or no path traversed data available, then HSGV goes into auto exploration process, which will be explained in later paragraphs. If the features are duplicate but no map database is found or if the extracted features are unique, they are saved with proper labelling such that the name of the camera which captured them and the current position of Vehicle can be retrieved.

[0101] After this Vehicle runs Depth Al algorithm on the captured images so that the distance data of every pixel in the image from the Vehicle is calculated. Depth output generate a colour gradient view of the image, with certain colour representing nearest pixel and a certain colour for the farthest pixel. In numerical view, it generates a matrix of pixel distance data for each image. Distance gradient between pixels will be small and similar value in the direction of smooth surfaces without obstructions. Wherever there are sharp changes in the scene, like edges of an object, or edges of floor and wall, the corresponding distance data and the colour will have a large gradient. HSGV estimates a threshold gradient value based on previous supervised learning. Wherever the pixel to pixel gradient is larger than the threshold, the coordinate of those pixels called edge pixels are saved and rest of the pixels are 0. In parallel, HSGV also runs object recognition function on the captured images and bounding boxes of objects are created. Dynamic and static objects are identified based on pretrained object recognition models. Bounding boxes of static and dynamic objects are preserved. This generates a matrix, such that bounding box pixels of dynamic objects are 0 and static objects are -1 and rest of the pixels are 1.

[0102] Now the output matrices of the two processes Depth and Object recognition are multiplied element-wise through Hadamard product. Whichever pixels detected dynamic object, their distance data becomes zero, whichever pixels detected static object, their distance data becomes negative, and distance data of all other edges remain unchanged. This new matrix is passed through a noise filter which multiplies the vertically lowest negative element in each column by -1 and by 0 for all other negative elements. The noise filter may also take input of obstacle sensors like 2D lidars, Ultrasonic sensors, ToF sensors which can also measure distance of the nearest obstacle around the HSGV in order to further update the matrix data. The matrices, thus generated by processing images from cameras in different directions are stitched together to create a 2D map with HSGV at its center. This map only consists of static objects’ projection on the floor, available free space on the workfloor while dynamic objects are ignored and re-mapped in the subsequent exploration process. Its recommended that dynamic objects are not present in the view of HSGV while training for more accurate mapping in a short time.

[0103] HSGV then analyzes the output map to determine which available free spaces are traversable such that width of the free spaces is greater than the rotation diameter of HSGV by a margin of 1 feet. This margin is tunable. Among the available free space, HSGVorients in the direction of the longest traversable free space and moves forward until the distance covered is equal to the estimated length of the free space or until an obstacle is detected, whichever happens first. If it’s a dynamic object, then it stops, raises different forms of alert to request the dynamic object to move and waits until the obstacle is no longer there. If it’s a static object, then take a turn such that there is no more static object in the front view of the HSGV and stop for the next round of capturing images and map creation. During this process, HSGV also updates the map with the path traversed by HSGV. This enables HSGV to cover all the available traversable free spaces and map the entire floor.

[0104] If the training happens outdoors, Vehicle first checks if lane detection is ON. If it is not ON, then the training process is exactly same. If it is ON, then only difference is that HSGV always follows lane markers instead of defining path for automatic exploration.

[0105] In an embodiment, map plotting implies creation of the 2-dimensional representation of the traversable space marked by coordinates and lines joining them. This is aided with uploading of an existing layout which directly gives the relationship of all the static obstacles and the boundaries (including walls and openings) in the form of coordinates and lines joining them. To get a more accurate environment map, which includes changes not reflected in the layout, addition of known static object locations, auto-exploration is performed.

[0106] Figure 5 illustrates the flow diagram of Autonomous Mobile Robots (AMR) Navigation Mode of the hybrid self-driving ground vehicle according to an exemplary implementation of the present invention.

[0107] The figure illustrates the flow diagram of Autonomous Mobile Robots (AMR) Navigation Mode of the hybrid self-driving ground vehicle. If the vehicle is configured as AMR, and at least one training data exists for at least one map (floor or campus layout) in the system, the user can switch to Navigation mode. In this mode, user can assign tasks to HSGV such as to navigate from one point to another within a trained space, automatically, with dynamic path planning without needing any fixed path training. The user can input a destination on the UI, select the floor map where HSGV is located and trigger the HSGV to move from its current position to the given destination. If no training data is present, user is instructed to go to training mode or select another map. This check allowsthe user to be sure if the selected map is correct and also enables user to train HSGV on the existing map if no previous training has happened.

[0108] Before HSGV can start moving towards the destination, it has to find its current position. It retrieves the last known position from its own memory and the local server. If both match, means HSGV was not moved from its last known position. If both do not match or if data from server is not available, HSGV last position is unknown. HSGV starts computing its current location by capturing the images of its surrounding through all the active cameras. It extracts the features of these images by using ORB / SIFT algorithms. The extracted features are compared with the feature database and if there is a match beyond a certain threshold, it is considered as duplicate. HSGV is able to identify its current position and updates the same as source. If it is not a match, then the steps mentioned in Paragraphs [0101 and 0102] are repeated in the same order. HSGV overlays the output map on the trained map until a match is found. Once maps are matched, current position is identified as source.

[0109] If map matching fails, then alerts are raised and control is transferred to the user. User then brings the HSGV in a trained area and then the process starts again from the paragraph

[0108] onwards.

[0110] Once source is updated, the vehicle checks if the source coordinates are same as that of the destination. If it is true, then HSGV has successfully arrived at the destination and is ready for another navigation mission. Else, it computes the shortest path from the source to the given destination from the available traversable free space as computed during training mode. This ensures that the shortest path is not obstructed by any wall or immovable static objects. Once shortest path is calculated, HSGV orients itself accordingly and starts moving towards the destination.

[0111] During this motion if vehicle detects any obstacle, it follows the steps mentioned in paragraph

[0092] . In case, the detected obstacle does not move away after a predefined wait time, then HSGV computes the a re-routing path around the obstacle to merge with the previously calculated shortest path. This is possible because of the pre-trained object database which includes the dimension of the objects. For any other unknown object, re-routing is done differently by moving away from the object and re-calculating the shortest path from there. HSGV keeps moving forward on the updated path until the current positioncoordinates become same as that of the destination. HSGV also computes its current position after every “predefined time delay” to calibrate its vision system with its odometry system. Current position computation is also performed when HSGV encounters obstacles and reaches the destination.

[0112] Whether in AGV or AMR navigation mode, HSGV has feed back (odometry) system which keeps it aligned with the computed path. Also, it updates its current position in the system as well as sends the data to the server through the connected wireless network. So, HSGV can be tracked live from any UI connected to the network.

[0113] In an embodiment, during navigation in AMR mode, first current position is localized and then the shortest path is planned from the available free spaces as recorded during the self-training. Re-routing process is initiated only when the obstacle detected is a static type and is performed through steps like estimating the dimension of the obstacle and re-running the shortest algorithm by creating a boundary around the obstacle (not explicitly mentioned in the flow diagram). Re-routing is generally not performed for dynamic obstacles like humans which can move when alerted. But it can be performed with a disable and enable option available on the user interface. It is clear that localization is performed through image matching and if failed, object-based localisation is performed with respect to multiple objects, through a process called trilateration. It is also implied that once localisation is done, the system performs shortest path planning to the destination based on a previously trained map. Mapping is not done continuously but only at selected locations like source, destination, obstacle detection points etc. which are also referred to as keyframes.

[0114] Figure 6 illustrates the depth function output of an image using the Depth Al models according to an exemplary implementation of the present invention.

[0115] The figure shows the output of processing an image using the Depth Al models. It creates a colour gradient pattern showcasing the distance of all the pixels in the image. Here, yellow shows the closest pixel and black the farthest.

[0116] Figure 7 illustrates the flow diagram of object training according to an exemplary implementation of the present invention.

[0117] The figure illustrates the flow diagram of object training. There are few Object Recognition Al models available in the art. These models are trained with the data of generic objects which are found in our day to day lives. But they may or may not have the data for the machines or any other objects which are commonly used within a warehouse, factory environment or any other specific facility who find the usefulness of self driving vehicle for logistics. Therefore, re-training the models to create a more upgraded model makes sure HSGV can make intelligent decisions by detecting the target objects or obstacles during motion.

[0118] During training (manual or automatic), selected static objects are visually captured from multiple viewpoints, recognized using trained object recognition models, associated with estimated dimensions and orientation and registered with a positional relationship relative to: a known reference point, a marker-aligned stop point, or the navigation-oriented environment model. The object’s spatial information is stored in a navigation-relevant form, sufficient to support later localization but not requiring full 3D surface reconstruction.

[0119] In one embodiment of the present disclosure / invention, this is done with the following steps.1. First select the object to be trained and place the vehicle at a known distance where the object is visible to HSGV i.e., SELF DRIVING GROUND VEHICLE. This can be verified using the camera view on HSGV dashboard.2. HSGV captures few images, annotates them using Mask-R CNN model. HSGV then rotates right side by a fixed angle such as 30 degrees and repeats the images capturing and annotation process until the object becomes invisible.3. HSGV is again positioned on the other sides of the object and then step 2 is repeated.In this way, around 50 iterations are done to collect object data from all directions.4. Then step 2 and 3 are repeated again by changing the lighting conditions if its indoors and at different time of the day like morning, noon, afternoon, evening and night if its outdoors. This is done for around 50 iterations. So, total more than 2000 images of the object are collected and annotated automatically.5. These annotations are manually verified and the result is given as a feedback to the model so that annotation model improve overtime.6. Then these correct annotations are divided into 70% and 30% parts called group 1 and group 2 respectively. Group 1 is used to train the YOLO v8 model for object recognition and the group 2 is used to test the efficacy of the newly trained model. Then the object identity is saved in the form of bounding boxes, confidence scores, class name, object dimensions etc.7. For dimension calculation, HSGV uses the known distance information and trigonometry to find the width, length and height of the object.8. The entire process is repeated for as many unique objects as available in the facility.And in this way, a powerful model is created which is then used by HSGV during training or navigation operations.Object Detection During Navigation

[0120] During navigation, the vehicle periodically performs object recognition using one or more monocular cameras. When a trained reference object is detected, the system: identifies the object class and instance, determines its relative position within the camera frame, estimates relative distance using monocular depth inference and computes the object’s bearing relative to the vehicle. This information provides a relative spatial constraint between the vehicle and the known object.

[0121] In one embodiment, when multiple reference objects are detected simultaneously or sequentially, the system uses their known spatial relationships to refine the vehicle’s position. As per figure 4 and 5, images are captured from the available cameras (plural), object recognition thread is run which can detect static objects (plural although not explicitly mentioned) and filter out dynamic objects. It plots the coordinates of the static objects, therefore enabling localisation with respect to multiple objects, which is done through map matching thread. Map matching can be successfully done with one or more objects. For example: relative distance and bearing to a first object constrains the vehicle to a region, relative distance and bearing to a second object further reduces uncertainty, detection of a third object resolves ambiguity. This process enables robust localization through geometric consistency, without requiring dense environmental mapping. The number of objects required may vary based on: environment structure, confidence thresholds, task accuracy requirements.

[0122] Object-based localization is particularly effective for correcting accumulated odometry drift. When the system detects sufficient object-based confidence, it: compares the expected relative position of objects with observed measurements, computes a correction for positional and / or angular error, updates the vehicle’s pose estimate accordingly. This correction is applied intermittently, typically at: navigation pauses, obstacle-induced stops, confidence degradation events and other keyframes.

[0123] Importantly, object-based localization: does not require continuous updating of the global map, does not require re-estimation of free space, does not require point cloud alignment. The navigation-oriented environment model remains stable, while object observations are used to validate and correct pose, not to rebuild the environment representation. The system is designed to operate under partial visibility conditions. If an object is partially occluded, or temporarily blocked by a dynamic entity, the system reduces confidence weighting, defers correction or relies on odometry until additional references become available. This prevents false localization updates and ensures stability in human-dense environments.

[0124] Object-based localization operates independently of, and complementary to, marker-based alignment. In environments where markers are: unavailable, damaged, or impractical, object-based localization provides anon-intrusive alternative. In environments where both are available, the system may: prefer marker-based alignment for absolute correction, use object-based localization for intermediate refinement.

[0125] Figure 8 illustrates the flow diagram of the concept of Keyframe-based Intermittent SLAM according to an exemplary implementation of the present invention.

[0126] The figure illustrates the flow diagram of the concept of Keyframe-based Intermittent SLAM. HSGV performs localization continuously with the help of odometry sensor system. But it performs mapping only when necessary. Mapping in AMR mode is generally done at critical points such as at the start and end of a navigation process, when an obstacle is encountered and re-routing is necessary. In AGV mode, mapping is done only at stop / virtual points. Therefore, simultaneous mapping and localization is only intermittent and not continuous and is based on critical frames or keyframes.

[0127] The present invention uniquely combines the capabilities of both AGVs and AMRs, allowing it to switch operational modes based on specific needs. This adaptabilityensures that whether indoors or outdoors, whether precise paths are necessary, or flexibility in pathfinding is important, the present invention can operate efficiently. This dual-mode capability ensures to adapt to both structured and dynamic environments, facilitating seamless transitions and reducing the need for multiple specialized vehicles.

[0128] In general, many current systems, particularly AGVs, require complex reprogramming to alter routes, which can be a significant barrier in dynamic industrial environments. The lack of simple training option limits the ability of operators to quickly adapt routes based on changing conditions or safety considerations.

[0129] In general, AMRs require an initial mapping of a workspace for referencebased positioning. Most AMRs have only automatic mapping in which AMRs move in a specified pattern on its own and keeps mapping until it reaches the first mapped area again. This may not be suitable when areas to be mapped are large as it consumes too much time. Also, it is redundant when there are few allowed paths / lanes. Some AMRs only have manual training / mapping in which a human pushes or pulls the vehicle on a certain path across the workspace. This process is highly inconvenient and extremely slow.

[0130] However, according to the present invention, the proposed system introduces dual training modes — Manual and Automatic. The Manual mode allows on-site training with a simple handheld wireless remote, making it accessible for non-technical personnel to define and modify routes as needed. This capability not only enhances operational flexibility but also significantly improves safety. Operators can designate specific lanes, avoiding areas with high human traffic and optimizing the workflow. This level of control is crucial in busy or changing environments, ensuring both safety and efficiency.

[0131] The Automatic mode employs an innovative method of creating maps from images captured at certain instances defined by unique algorithm. This is different from the 3D camera-based SLAM in which a video stream is analyzed continuously to map an environment. Present invention includes auto labeling and mapping of objects those are part of the environment and intelligently ignoring the dynamic obstacles. This makes the mapping highly accurate, less data-intensive and easier to update. This process not only simplifies the operational complexity but also reduces the dependence on specialized training or technical skills.

[0132] According to the present invention the manual training mode empowers operators with the ability to easily define and modify navigation paths using a simple hand-held remote. The automated training mode empowers users to sit-back in their room and watch it map an entire area on its own through their computing device screen.

[0133] Most AMRs require continuous 3D mapping whether mapping a work-floor for the first time or running a navigation mission on a mapped work-floor. These detailed complex 3D mapping requires significant data storage and processing power, which may not be feasible for many industrial applications. These systems can also be less intuitive for operators to use and understand, potentially hindering their effective adoption.

[0134] The present invention employs keyframe base intermittent SLAM using 2D cameras. In this process, simplified 2D maps are created during training period, making it easier for users to understand and interact with the system. These 2D maps in combination with odometry feedback are then used for live vehicle tracking and route selection, ensuring that the system remains user-friendly and accessible. Since, mapping is done only once in AGV mode and only at selected few keyframes in AMR mode, this approach reduces data complexity and storage requirements while providing the necessary functionality for efficient vehicle navigation. This process reduces the average load on network bandwidth to 8Mbps from 80Mbps in 3D lidar scenario and 500Mbps in 3D camera-based scenario. This approach reduces the cognitive load on users making the technology more accessible and less costly to maintain.

[0135] In general, available solutions use coordinate-based inputs for source and destination, which can be inconvenient, difficult to remember and impractical. For instance, instructing a robot to move to coordinates (100, 120) is far less intuitive than directing it to "Machine A."

[0136] The present invention has simplified the navigation process by allowing operators to assign recognizable names to destinations. This method is far more intuitive and reduces errors associated with coordinate systems. Whether directing to "Shipping Dock" or "Assembly Line 3," operators can easily select destinations name instead of typing in (100,120), or (2594, 6987), making the system more user-friendly and efficient. This feature is particularly beneficial in large facilities where navigation can otherwise be complex.

[0137] Figure 9 illustrates a method for autonomous navigation of a hybrid selfdriving ground vehicle according to an exemplary implementation of the present invention.

[0138] The figure 9 illustrates a method for autonomous navigation of a hybrid selfdriving ground vehicle. The method comprises actuating (910) movement of the vehicle viaa mobility platform, continuously (920) estimating translational and rotational motion of the mobility platform using an odometry sensor system, acquiring (930) image data of an environment with a visual sensing system comprising a plurality of two-dimensional cameras, detecting (940) dynamic and static obstacles using an obstacle detection system, selecting (950) between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode, continuously (960) localizing the vehicle using the odometry sensor system, intermittently (970) performing environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment and autonomously (980) navigating the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

[0139] The present method in AGV mode, the method comprises computing a navigation path along a user-trained path defined by a plurality of stop points and vector relationships between the stop points, detecting and avoid collision with the obstacles detected by the obstacle detection sensor system and realigning position and orientation of the vehicle at the stop points using at least one of a physical marker, a pre-trained static object, or a virtual reference point upon user confirmation, wherein the realignment resets accumulated positional and orientation error.

[0140] In another embodiment, in the AGV mode, the method further comprises bringing the vehicle to a standstill upon detecting an obstacle in the pre-trained fixed path and generating alerts through at least one of visual indicators, audible signals, or notifications to a supervisory system, without performing autonomous rerouting.

[0141] The present method in the AMR mode, the method comprises computing a navigation path on a self-trained path to a user-specified semantic destination over the two-dimensional navigation-oriented representation; detecting and avoiding obstacles detected by the obstacle detection system and rerouting around detected obstacles with intermittent calibration without performing full environment remapping.

[0142] In another embodiment, in the AMR mode, the method comprises computing a bypass trajectory around a detected obstacle based on at least one of pre-trained object dimensions or newly detected obstacle geometry and merging the bypass trajectory with a previously computed path.

[0143] In the present invention, acquiring image data comprises selectively activating a front-facing camera, a rear-facing camera, left and right side-facing cameras, and a downward-facing camera based on user application, type of environment, the operational mode and selected keyframe requirements. The selected keyframes comprise at least one of navigation initiation, arrival at a stop point, arrival at a virtual reference point, detection and alignment with a marker, recognition of a pre-trained static object, obstacle-induced stopping, rerouting events, or completion of navigation.

[0144] The method further comprises performing object-based localization by recognizing one or more pre-trained static objects and using spatial relationships between the vehicle and the recognized objects to correct odometry drift or refine vehicle pose, without performing continuous global environment remapping.

[0145] In the present invention, executing the method comprises operating a primary processor to perform navigation mode management, task planning, user interface handling, and keyframe mapping control and operating at least one secondary processor to perform mobility control, safety monitoring, sensor preprocessing, and external subsystem interfacing.

[0146] The method further comprises triggering a controlled stop using a safety sensor system including at least one of an emergency stop switch, a collision sensor, or a tilt sensor upon detection of unsafe conditions or loss of localization.

[0147] The method further comprises automatically triggering actions through external mechanical or electronic modules selected from conveyors, lifting mechanisms, robotic arms, pallet handling systems, inspection or sensing modules upon reaching designated stop points via an external subsystem interface.

[0148] The method further comprises dynamically adjusting selection between the Automated Guided Vehicle operational mode and the Autonomous Mobile Robot operational mode based on task requirements and environmental conditions.

[0149] The present invention revolutionizes various sectors by automating material transportation and enhancing operational efficiency. In manufacturing and industrial facilities, the system autonomously transports materials between different stages of theproduction process, from raw materials reception to processing and assembly lines. It optimizes the flow of semi-finished goods within the facility, reducing handling time and improving efficiency. The present invention also ensures the timely movement of finished goods from assembly lines to packaging and dispatch areas, enhancing overall productivity.

[0150] In warehousing and logistics, the present invention streamlines operations by assisting with order picking, retrieving goods from various parts of a warehouse according to order requirements. It ensures continuous inventory availability by automatically transporting inventory to areas with low stocks. Additionally, the system can be deployed for loading and unloading goods onto transport vehicles, reducing manual labor and expediting the process. In healthcare and hospital settings, it handles the transport of medical supplies, laboratory samples, and medical equipment within hospital complexes, ensuring timely and safe delivery. The system also aids in the movement of laundry and medical waste to designated areas, minimizing human handling and enhancing hygiene standards.

[0151] In one of the embodiments of the present alternative navigation sensors such as Hybrid Navigation Systems and Multimodal sensor integration can be used. For hybrid Navigation Systems by incorporating additional navigation technologies such as GPS for outdoor use and ultrasonic sensors for improved obstacle detection could be used. For multimodal Sensor Integration by incorporating thermal imaging, or more advanced depthsensing technologies, as one of the embodiments the present invention can operate more effectively in diverse conditions, such as in darkness, through smoke, or in visually obstructed environments.

[0152] In other embodiments of the present invention for enhancing communication capabilities 5G Connectivity can be used. Incorporating 5G technology could improve data transmission speeds and reliability, enabling more robust remote control and monitoring capabilities, particularly useful for coordinating large fleets.

[0153] In other embodiments of the present invention for alternative power solutions Solar Power Integration or automated charging can be used.

[0154] For operations in outdoor environments, integrating solar panels could help extend battery life and reduce the frequency of recharges needed, enhancing sustainability Solar Power can be integrated.

[0155] Implementing automated charging capabilities could allow HSGV to recharge itself whenever battery is low or it is in idle state, increasing operational uptime.

[0156] In other embodiments by modular attachments i.e. the ability to attach different modules for specific tasks (like carrying, lifting, quality and quantity inspection or surveillance) could make it more versatile and customizable according to user needs.Advantages:• Use of monocular camera (2D camera) as the main navigation sensor as opposed to the mandatory need of 3D lidar or 3D camera navigation in other autonomous vehicles. • Dual Mode operation: AGV and AMR modes.• Operation as AGV without the need for tracks, magnetic tapes etc.• Simplified Manual and Automatic Training: Most AGVs have no training involved only preprogramming which require skilled professional. Other AMRs have only automated training / mapping or only fully manual training / mapping. According to the present invention it has both manual with wireless remote and automated training so that it can suit different use cases.• Keyframe based intermittent Simultaneous Localization and Mapping: The present invention does not use continuous mapping. It uses an odometry system to compute the distance and angle moved and performs image-based mapping only at critical points or keyframes. In this process, images are processed using depth CNN + object recognition CNN models to identify available traversable free space, trained path and to calculate the position vectors of static objects, to create a 2D map. During a navigation mission in MAR mode, the proposed system computes its position with respect to the created map only 2 times + each time there is a re-routing situation due to obstacle. Even in AGV mode navigation, mapping is done only at stop / virtual points. Therefore, computation burden is drastically reduced while the vehicle is tracked continuously live with the help of odometry system.• User-Friendly Navigation Input: Most autonomous systems use destination coordinates for navigation, which can be cumbersome and error-prone, especially in large facilities. The proposed system automatically labels identified objects and markers which can be re-labeled by the user on the go or later. This enables users to input these labels as source and destination for navigation. Also, there is a provision to track an object which is suitable for many use cases.

[0157] The various embodiments described above are specific examples of a single broader invention. Any modifications, alterations or the equivalents of the above-mentionedembodiments pertain to the same invention as long as they are not falling beyond the scope of the invention as defined by the appended claims. It will be apparent to a person skilled in the art that the hybrid self-driving ground vehicle and a method thereof may be provided using some or many of the above-mentioned features or components without departing from the scope of the invention. It will be also apparent to a skilled person that the embodiments described above are specific examples of a single broader invention which may have greater scope than any of the singular descriptions taught. There may be many alterations made in the invention without departing from the spirit and scope of the invention.

[0158] Figures are merely representational and are not drawn to scale. Certain portions thereof may be exaggerated, while others may be minimized. Figures illustrate various embodiments of the invention that can be understood and appropriately carried out by those of ordinary skill in the art.

[0159] In the foregoing detailed description of embodiments of the invention, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed embodiments of the invention require more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby incorporated into the detailed description of embodiments of the invention, with each claim standing on its own as a separate embodiment.

[0160] It is understood that the above description is intended to be illustrative, and not restrictive. It is intended to cover all alternatives, modifications and equivalents as may be included within the spirit and scope of the invention as defined in the appended claims. Many other embodiments will be apparent to those of skill in the art upon reviewing the above description. The scope of the invention should, therefore, be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. In the appended claims, the terms “including” and “in which” are used as the plain-English equivalents of the respective terms “comprising” and “wherein,” respectively.

Claims

Claims :

1. A hybrid self-driving ground vehicle (100) for autonomous navigation, the hybrid self-driving ground vehicle (100) comprising:a mobility platform (102) configured to actuate movement of the vehicle;an odometry sensor system (104) configured to continuously estimate translational and rotational motion of the mobility platform;a visual sensing system (106) comprising plurality of two-dimensional camera configured to acquire image data of an environment;an obstacle detection system (108) configured to detect dynamic and static obstacles; one or more processors (110, 112) operatively coupled to the odometry sensor system, the visual sensing system, and the obstacle detection system, wherein the one or more processors configured to:select between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode;continuously localize the vehicle using the odometry sensor system; intermittently perform environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment; andautonomously navigate the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

2. The hybrid self-driving ground vehicle as claimed in claim 1, wherein in the AGV mode, one or more processors are configured to:compute a navigation plan along a user-trained path defined by a plurality of stop points and vector relationships between the stop points;detect and avoid collision with the obstacles detected by the obstacle detection sensor system; andrealign position and orientation of the vehicle at the stop points using at least one of a physical marker, a pre-trained static object, or a virtual reference point upon user confirmation, andwherein the realignment resets accumulated positional and orientation error.

3. The hybrid self-driving ground vehicle as claimed in claim 1, wherein in the AMR mode, one or more processors are configured to:compute a navigation path on a self-trained path to a user-specified semantic destination over the two-dimensional navigation-oriented representation;detect and avoid obstacles detected by the obstacle detection sensor system; and reroute around detected obstacles with intermittent calibration without performing full environment remapping.

4. The hybrid self-driving ground vehicle as claimed in claim 1, wherein the plurality of two-dimensional camera comprises a front-facing camera, a rear-facing camera, left and right side-facing cameras, and a downward-facing camera, and wherein camera activation is selectively controlled based on user application, type of environment, the operational mode (AGV or AMR) and selected keyframe requirements.

5. The hybrid self-driving ground vehicle as claimed in claim 1, wherein the selected keyframes comprise at least one of home position, navigation initiation, arrival at a user defined stop point, arrival at a virtual reference point, detection and alignment with a fiducial marker, recognition of a pre-trained static object, obstacle -induced stopping, rerouting events, or completion of navigation.

6. The hybrid self-driving ground vehicle as claimed in claim 1, wherein the navigation-oriented representation comprises representation of static obstacles, traversable free space, semantic destinations, and reference relationships, without storing a dense geometric map.

7. The hybrid self-driving ground vehicle as claimed in claim 1, wherein the one or more processors are further configured to perform object-based localization by using one or more pre-trained static objects as spatial reference anchors to determine or correct the position and orientation of the vehicle independent of physical markers or continuous environment mapping.

8. The hybrid self-driving ground vehicle as claimed in claim 2, wherein in the Automated Guided Vehicle (AGV) mode, the vehicle follows a pre-trained fixed path and comes to a standstill upon detecting an obstacle in the pre-trained path and generate alerts through at least one of visual indicators, audible signals, or notifications to a supervisory system, without performing autonomous re-routing.

9. The hybrid self-driving ground vehicle as claimed in claim 2, wherein the Automated Guided Vehicle (AGV) mode comprises manual training of navigation paths using a handheld remote, teach-pendant-like interface, or user interface.

10. The hybrid self-driving ground vehicle as claimed in claim 1, wherein markers are used only at selected semantic locations and not in a uniform grid or markers are not used at uniform distance on a line.

11. The hybrid self-driving ground vehicle as claimed in claim 3, wherein in the AMR mode, the processor is configured to compute a bypass trajectory around a detected obstacle based on at least one of pre-trained object dimensions or newly detected obstacle geometry, and merge with a previously computed path.

12. The hybrid self-driving ground vehicle as claimed in claim 1, wherein one or more processors comprise a primary processor executing navigation mode management, task planning, user interface handling, and keyframe mapping control, and at least one secondary processor executing mobility control, safety monitoring, sensor preprocessing, and external subsystem interfacing.

13. The hybrid self-driving ground vehicle as claimed in claim 5 or 12, wherein camera activation and processing frequency are adaptively / dynamically adjusted based on operational mode, environmental complexity, or task criticality.

14. The hybrid self-driving ground vehicle as claimed in claim 1, further comprising a safety sensor system (114) including at least one of emergency stop switch, a collision sensor, or a tilt sensor, wherein the safety sensor system is configured to trigger a controlled stop upon detection of unsafe conditions or loss of localization.

15. The hybrid self-driving ground vehicle as claimed in claim 1, further comprising an external subsystem interface (116) configured to automatically trigger actions through external mechanical or electronic modules selected from conveyors, lifting mechanisms, robotic arms, pallet handling systems, inspection or sensing modules upon reaching designated stop points.

16. The hybrid self-driving ground vehicle as claimed in claim 1, wherein the selection between the Automated Guided Vehicle mode and the Autonomous Mobile Robot mode is adjusted dynamically based on task requirements and environmental conditions.

17. A method (900) for autonomous navigation of a hybrid self-driving ground vehicle, the method comprising:actuating (910) movement of the vehicle via a mobility platform;continuously (920) estimating translational and rotational motion of the mobility platform using an odometry sensor system;acquiring (930) image data of an environment with a visual sensing system comprising a plurality of two-dimensional cameras;detecting (940) dynamic and static obstacles using an obstacle detection system; selecting (950) between an Automated Guided Vehicle (AGV) operational mode and an Autonomous Mobile Robot (AMR) operational mode;continuously (960) localizing the vehicle using the odometry sensor system; intermittently (970) performing environment mapping at selected keyframes by processing image data from the two-dimensional cameras to generate and update a two-dimensional navigation-oriented representation of the environment; and autonomously (980) navigating the vehicle within the environment while maintaining positional accuracy through intermittent calibration.

18. The method as claimed in claim 17, wherein in the AGV mode, the method comprises computing a navigation path along a user-trained path defined by a plurality of stop points and vector relationships between the stop points;detecting and avoid collision with the obstacles detected by the obstacle detection sensor system; andrealigning position and orientation of the vehicle at the stop points using at least one of a physical marker, a pre-trained static object, or a virtual reference point upon user confirmation, wherein the realignment resets accumulated positional and orientation error.

19. The method as claimed in claim 17, wherein in the AMR mode, the method comprisescomputing a navigation path on a self-trained path to a user-specified semantic destination over the two-dimensional navigation-oriented representation;detecting and avoiding obstacles detected by the obstacle detection system; and rerouting around detected obstacles with intermittent calibration without performing full environment remapping.

20. The method as claimed in claim 17, wherein acquiring image data comprises selectively activating a front-facing camera, a rear-facing camera, left and right side-facing cameras, and a downward-facing camera based on user application, type of environment, the operational mode and selected keyframe requirements.

21. The method as claimed in claim 17, wherein the selected keyframes comprise at least one of home position, navigation initiation, arrival at a stop point, arrival at a virtual reference point, detection and alignment with a marker, recognition of a pre-trained static object, obstacle-induced stopping, rerouting events, or completion of navigation.

22. The method as claimed in claim 17, further comprising performing object-based localization by recognizing one or more pre-trained static objects and using spatial relationships between the vehicle and the recognized objects to correct odometry drift or refine vehicle pose, without performing continuous global environment remapping.

23. The method as claimed in claim 18, wherein in the AGV mode, the method further comprises bringing the vehicle to a standstill upon detecting an obstacle in the pre-trained fixed path and generating alerts through at least one of visual indicators, audible signals, or notifications to a supervisory system, without performing autonomous rerouting.

24. The method as claimed in claim 19, wherein in the AMR mode, the method comprises computing a bypass trajectory around a detected obstacle based on at least one of pre-trained object dimensions or newly detected obstacle geometry and merging the bypass trajectory with a previously computed path.

25. The method as claimed in claim 17, wherein executing the method comprises operating a primary processor to perform navigation mode management, task planning, user interface handling, and keyframe mapping control and operating at least one secondary processor to perform mobility control, safety monitoring, sensor preprocessing, and external subsystem interfacing.

26. The method as claimed in claim 17, further comprising triggering a controlled stop using a safety sensor system including at least one of an emergency stop switch, a collision sensor, or a tilt sensor upon detection of unsafe conditions or loss of localization.

27. The method as claimed in claim 17, further comprising automatically triggering actions through external mechanical or electronic modules selected from conveyors, lifting mechanisms, robotic arms, pallet handling systems, inspection or sensing modules upon reaching designated stop points via an external subsystem interface.

28. The method as claimed in claim 17, further comprising dynamically adjusting selection between the Automated Guided Vehicle operational mode and the Autonomous Mobile Robot operational mode based on task requirements and environmental conditions.

Images