A multi-mode switched autonomous navigation and obstacle avoidance method and system for a mars rover

Abstract

The application relates to the technical field of deep space exploration and autonomous control, and provides a multi-mode switching Mars exploration vehicle autonomous navigation obstacle avoidance method and system, which comprises the following steps: S1: calculating the environment complexity level of the current road condition based on environment perception data; S2: selecting a current working mode from a preset full autonomous mode, a semi-autonomous mode and a complete program control mode according to the environment complexity level; S3: in the confirmed or corrected working mode, a travel task is executed according to the control authority distribution strategy corresponding to the mode, and corresponding level autonomous obstacle avoidance decision is executed based on the obstacle detection and ranging results in the travel process; and S4: after completing the travel task of the current stage, a hierarchical energy saving operation is automatically executed, and the travel task parameters of the next stage are dynamically adjusted according to the energy state prediction result. The technical scheme can flexibly switch the working mode according to the road condition complexity, can give consideration to autonomous obstacle avoidance and ground program control, and has an automatic energy saving management capability.

Description

Technical Field

[0001] This invention relates to the technical field of deep space exploration and autonomous control, and particularly to a multi-mode switching method and system for autonomous navigation and obstacle avoidance of a Mars rover. It is applicable to Mars rovers flexibly switching operating modes according to different road conditions in complex terrain environments, achieving coordinated operation of autonomous navigation and obstacle avoidance, semi-autonomous navigation, and remote programmable control. Background Technology

[0002] Mars, as one of the planets in the solar system most similar to Earth in environment, has always been a key target for deep space exploration. With the continuous advancement of Mars exploration missions by various countries, the ability of rovers to move and operate on the Martian surface directly determines the breadth and depth of scientific exploration. However, the Martian surface has extremely complex terrain, widely distributed with impact craters, craters, canyons, steep slopes, and gravel features. These complex terrain areas contain the richest scientific information and are high-value exploration points of great interest to geologists and planetary scientists. However, it is precisely these rugged terrain conditions that place extremely stringent demands on the autonomous driving capabilities of Mars rovers. During on-site operation, the rover must be able to identify and avoid various obstacles in real time. Any slight operational error could lead to vehicle rollover, suspension system damage, or wheel jamming, thus causing the entire exploration mission to terminate prematurely and resulting in irreparable scientific losses.

[0003] Regarding communication, the greatest distance between Mars and Earth can reach approximately 400 million kilometers, with radio signal one-way transmission delays reaching up to 22.22 minutes. This means that after ground operators send a control command, they need to wait nearly 45 minutes to receive feedback from the rover on its execution results, making real-time control impossible. Under such high-latency communication constraints, relying entirely on ground remote control to guide every step of the rover's movement is not only extremely inefficient, but also often leads to the rover stalling or even becoming dangerous when encountering sudden obstacles or terrain changes, as it cannot receive timely ground commands. Therefore, endowing the rover with a certain degree of autonomous navigation and obstacle avoidance capability is a fundamental prerequisite for carrying out missions on the Martian surface.

[0004] On the other hand, in certain extremely complex terrain environments—such as steep slopes covered with sharp rocks, deep pits with soft edges, or canyons with obstructed visibility—the rover's built-in autonomous obstacle avoidance algorithm may misjudge due to noise in the perception data, terrain modeling biases, or insufficient strategy generalization, leading the vehicle to choose the wrong detour path or misjudge the safety of passage. In such high-risk scenarios, the experience and global judgment of ground experts remain indispensable, requiring precise intervention in the rover's actions through remote control. Thus, the design of the Mars rover's control system faces a dilemma: over-reliance on autonomy can lead to uncontrollable extreme risks, while over-reliance on remote control cannot cope with long-term time-delay constraints.

[0005] Currently, existing Mars rover control systems generally only have two operating modes: fully autonomous mode and fully remote control mode. In fully autonomous mode, the rover relies entirely on its onboard sensors (such as stereo cameras and inertial measurement units) and built-in navigation and obstacle avoidance algorithms for path planning and obstacle avoidance, requiring no real-time ground intervention. This mode is suitable for relatively open and regular terrain. However, when facing extremely complex terrain, this mode is prone to misjudgment or failure due to the inherent limitations of the algorithm's perception and decision-making capabilities. In fully remote control mode, the rover strictly follows every action command sent from the ground. All movement decisions are made by ground personnel based on transmitted images and telemetry data. While this mode can fully utilize the judgment of human experts, it is severely limited by communication latency. Each control action requires a long wait and has extremely high requirements for the continuity and bandwidth of the communication link, making it difficult to execute efficiently in actual missions.

[0006] It is evident that the existing two-mode solutions lack a smooth transition between "autonomous capability" and "human intervention." The autonomous mode is suitable for simple road conditions, while the remote control mode is suitable for extreme risk scenarios. However, there are numerous moderately complex road conditions between the two, requiring both macro-level directional guidance from ground experts and flexible autonomous obstacle avoidance capabilities from the rover during its operation. Current technologies do not provide such an intermediate working mode that can flexibly switch according to the complexity of road conditions, thus failing to balance the global decision-making advantages of ground experts with the local real-time response advantages of the rover. This gap in modes forces the operating team to make difficult choices between two extreme modes in actual exploration missions: either subjecting the rover to the risk of autonomous misjudgment or sacrificing operational efficiency for long delays, making it difficult to achieve optimal mission execution.

[0007] Furthermore, the Mars rover's energy supply relies entirely on solar panels or nuclear batteries, with extremely limited power output. Environmental factors such as Martian surface lighting conditions and dust storms further exacerbate the uncertainty of energy acquisition. In non-operational states, if the rover continuously keeps its cameras on, its processor running at full speed, or its communication modules in high-power standby mode, it will rapidly consume precious electrical energy, significantly shortening the rover's effective operating days. Existing control systems generally lack automated energy-saving management mechanisms for mission downtime, typically requiring ground personnel to manually send hibernation commands or rely on timers to shut down. This approach increases the burden on ground operations and fails to allow for timely and autonomous entry into low-power modes based on mission completion status, resulting in unnecessary energy waste.

[0008] Based on the above analysis, the main shortcomings of the existing technology can be summarized as follows: First, the operating mode settings are unreasonable. The existing system only has a fully autonomous mode and a fully remote control mode, lacking an intermediate transition mode that can flexibly switch according to the complexity of road conditions, and cannot achieve on-demand dynamic adjustment between autonomous navigation and ground-based program control.

[0009] Second, there is a disconnect between autonomous decision-making and human intervention capabilities. Existing solutions cannot simultaneously consider the autonomous obstacle avoidance capabilities of the probe vehicle and the global decision-making capabilities of ground experts. In moderately complex road conditions, it is impossible to fully rely on autonomous algorithms or efficiently utilize human experience for macro-level guidance, making it difficult to simultaneously ensure mission execution efficiency and safety.

[0010] Third, energy management relies on manual operation. Existing control systems generally lack automated energy-saving management functions, and the probe cannot autonomously enter a low-power mode when not in operation or during mission intervals, resulting in unnecessary energy consumption and limiting the overall duration of the probe mission.

[0011] Therefore, there is an urgent need to provide a control method for Mars rovers that can flexibly switch working modes according to the complexity of road conditions, enabling them to switch between fully autonomous navigation, semi-autonomous navigation and fully programmed control modes as needed in different terrain environments. This method should take into account both the rover's autonomous obstacle avoidance capabilities and the global decision-making capabilities of ground experts, while also achieving automatic energy-saving management and extending the duration of the exploration mission, thus overcoming many shortcomings of the existing technologies. Summary of the Invention

[0012] To address the aforementioned problems, the present invention aims to provide a multi-mode switching autonomous navigation and obstacle avoidance method and system for Mars rover. It can flexibly switch operating modes according to the complexity of road conditions, balance autonomous obstacle avoidance with ground-based programmable control, and possess automatic energy-saving management capabilities.

[0013] The above-mentioned objective of this invention is achieved through the following technical solutions: A multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers includes: S1: Acquire environmental perception data around the detection vehicle, and calculate the environmental complexity level of the current road condition based on the environmental perception data; S2: Select the current working mode from the preset fully autonomous mode, semi-autonomous mode and fully programmed mode according to the environmental complexity level, and confirm or override the selected working mode through the mode identifier field of the ground programmed command. S3: In the confirmed or modified working mode, the vehicle performs the travel task according to the control authority allocation strategy corresponding to the mode, and performs the corresponding level of autonomous obstacle avoidance decision based on obstacle detection and ranging results during the travel. In the fully autonomous mode, the probe vehicle makes completely autonomous decisions and performs navigation and obstacle avoidance. In the semi-autonomous mode, the ground specifies macroscopic target parameters and the probe vehicle autonomously performs local obstacle avoidance. In the fully programmable mode, the ground completely takes over the control and the autonomous obstacle avoidance decision output is forcibly blocked. S4: After completing the current stage of the travel task, automatically perform graded energy-saving operations and dynamically adjust the parameters of the next stage of the travel task based on the energy status prediction results.

[0014] Further, in step S1, environmental perception data around the detection vehicle is acquired, and the environmental complexity level of the current road condition is calculated based on the environmental perception data, specifically as follows: The environmental perception data is collected by a binocular camera mounted on the detection vehicle. The environmental complexity level is calculated by weighted fusion of feature parameters from three dimensions: obstacle density, terrain undulation, and feasible area ratio. After the probe is powered on, it enters the fully autonomous mode by default. The central controller automatically generates a mode switching suggestion based on the real-time calculated environmental complexity level and sends a confirmation request to the ground through the wireless communication module. The ground completes the final confirmation or rejection of the mode switching by returning a confirmation command or a override command. When communication is interrupted and the environmental complexity level exceeds the emergency threshold, the central controller automatically performs a mode switch and transmits the switch record back to the ground via a low-power beacon.

[0015] Furthermore, in step S3, the control permission allocation strategy varies depending on the current working mode, specifically as follows: When the current working mode, after confirmation or correction, is the fully autonomous mode, the rover makes decisions entirely autonomously. It uses the obstacle detection module to detect obstacles on the road in real time, and the obstacle ranging module to calculate the three-dimensional coordinates and obstacle avoidance parameters of the obstacles. The autonomous obstacle avoidance module controls the rover to complete navigation and obstacle avoidance according to the obstacle avoidance strategy. After the rover travels a preset distance autonomously, it stops in place, turns off the camera, and enters the energy-saving mode. It continues to travel the preset distance on the next Martian day and then enters the energy-saving mode again. This cycle is repeated and is suitable for conventional exploration routes. The rover relies entirely on its own capabilities for navigation. When the current working mode, after confirmation or correction, is the semi-autonomous mode, the ground specifies the macroscopic task parameters, the probe vehicle autonomously performs local obstacle avoidance, the probe vehicle receives ground program control instructions, autonomously travels a specified distance from the starting point to the target point, performs autonomous obstacle avoidance function during the journey, and stops in place and turns off the camera after the instruction is completed, entering energy-saving mode. This mode is suitable for scenarios where ground experts specify the target direction and travel distance in moderately complex road conditions, and the probe vehicle autonomously completes obstacle avoidance. When the current working mode, after confirmation or correction, is the fully controlled mode, the ground takes complete control and shields the autonomous obstacle avoidance decision output. The probe strictly follows the ground control instructions to rotate the front of the vehicle by a specified angle and move forward a specified distance. During the journey, the obstacle avoidance function is not executed. After the instructions are executed, the probe stops in place and turns off the camera to enter energy-saving mode. This mode is suitable for scenarios where ground experts take complete control in extremely complex environments to prevent misjudgments by the autonomous obstacle avoidance algorithm. In the fully controlled mode, the obstacle avoidance decision output of the autonomous obstacle avoidance module is forcibly shielded during the journey. The execution of the current instructions is only paused when the single-line lidar triggers emergency braking, and execution is resumed after further instructions from the ground.

[0016] Further, in step S3, the execution of the corresponding level of autonomous obstacle avoidance decision based on the obstacle detection and ranging results includes an obstacle detection and ranging method, specifically: The autonomous obstacle avoidance decision-making utilizes a deep learning-based target detection algorithm to detect craters, rocks, and steep slopes on the Martian surface in real time. It uses a binocular stereo vision measurement algorithm to calculate the three-dimensional coordinates of the geometric center point of the obstacle and the three-dimensional coordinates of the four vertices of the circumscribed rectangle. Based on the three-dimensional coordinates, it calculates four obstacle avoidance parameters: azimuth, distance, width, and length of the obstacle. Based on the obstacle avoidance parameters and a preset obstacle avoidance strategy, it controls the differential speed between the left and right sides of the tracked chassis to achieve steering and obstacle avoidance on the rugged terrain of Mars.

[0017] Furthermore, the preset obstacle avoidance strategy adopts a four-level progressive obstacle avoidance decision-making mechanism, and the obstacle avoidance turning judgment includes a dynamic safety margin, specifically: The four-level progressive obstacle avoidance decision-making mechanism executes differentiated obstacle avoidance responses based on the characteristics of different obstacles on the Martian surface: the first level is the safe passage level, which maintains the original direction of travel when sufficient passage space is detected; the second level is the mild detour level, which calculates the minimum safe turning angle to avoid obstacles when they partially block the path; the third level is the tentative passage level, which approaches the obstacle at a speed lower than the normal travel speed and continuously evaluates it when the obstacle boundary is unclear or the risk is within a preset controllable range, and retreats to a safe position if the risk increases; the fourth level is the return route level, which automatically plans the return route when it is determined that the path ahead is completely impassable. When the distance to an obstacle is less than a preset threshold, obstacle avoidance judgment is initiated. The obstacle is determined to be located to the left, right, or directly in front of the probe based on its azimuth angle. The angle that the front of the probe needs to rotate is calculated based on the width and length of the obstacle and the cross-sectional dimensions of the probe. The probe is then controlled to rotate in place by the left and right tracks at a differential speed to avoid the obstacle. The calculation of the rotation angle includes a dynamic safety margin. The dynamic safety margin is adaptively adjusted according to the current speed of the probe, the friction coefficient of the Martian surface, and the type of obstacle. On typical soft sand or gravel surfaces on Mars, the safety margin is automatically increased to prevent sideslip. The autonomous obstacle avoidance decision-making also includes obstacle avoidance confidence assessment, outputting a confidence score for each obstacle avoidance decision. When the confidence score is lower than a preset low confidence threshold, the probe automatically decelerates to a safe speed and stops moving forward. At the same time, it requests instructions from the ground via the wireless communication module and maintains its current position and a minimum power consumption listening state while waiting for instructions from the ground. When the confidence score is higher than the preset low confidence threshold, the probe maintains the current obstacle avoidance decision and continues to execute.

[0018] Furthermore, in step S3, the detection vehicle is equipped with a binocular camera and a single-line lidar to form a heterogeneous fusion perception system, specifically: The binocular camera is used as the main environmental perception sensor to perform obstacle detection, ranging and terrain modeling on the Martian surface. The single-line lidar is used as an independent safety redundancy channel to trigger emergency braking when the distance between the rover and the obstacle exceeds the preset safety warning line, forming a two-layer protection architecture of binocular camera main perception and lidar safety redundancy. The heterogeneous fusion sensing system is equipped with a cross-validation mechanism for sensing confidence and a redundancy switching mechanism: when the binocular camera's detection confidence drops below a preset cross-validation threshold due to strong changes in lighting on the Martian surface, dust obstruction, or loss of texture features, the system automatically marks the low-confidence area and triggers the measurement data of the single-line lidar to cross-validate the low-confidence area. The system uses the lidar's ranging data to correct the binocular vision's perception results, compensating for blind spots and failure scenarios in visual perception under special lighting conditions on the Martian surface. When the binocular camera experiences a hardware failure or the sensing data remains abnormal, the system degrades to using the ranging data from the single-line lidar to maintain basic emergency braking functionality.

[0019] Furthermore, in step S4, the automatic execution of graded energy-saving operations and the dynamic adjustment of the travel task parameters for the next stage based on the energy state prediction results specifically involve: The graded energy-saving operation includes three levels: Level 1 energy saving, Level 2 energy saving, and Level 3 energy saving. The first level of energy saving is mild energy saving, which includes reducing the sampling frame rate of the binocular camera and reducing the transmission power of the wireless communication module; The second level of energy saving is medium energy saving, which includes turning off the single-line lidar and reducing the frequency of the central processing unit; The third level of energy saving is deep hibernation, which includes shutting down all unnecessary peripherals and switching the central controller to the lowest power standby state, retaining only the timed wake-up and communication monitoring functions; The rover automatically selects the appropriate energy-saving level based on the current mission stage and energy status in different working modes. After completing the mission instructions for the current stage in any mode, the rover automatically performs energy-saving operations: stops in place, turns off the camera, reduces the power consumption of the central controller, and waits for the next round of instructions or the next Martian day. The graded energy-saving operation is also equipped with an energy prediction model. The energy prediction model predicts the energy supply trend in the future time period based on solar panel output current monitoring data, battery remaining power data and historical data on Martian surface illumination changes. Based on the energy supply trend and current mission priority, the energy-saving level and mission execution plan are dynamically adjusted. High-priority scientific exploration missions are given priority in ensuring energy supply, while low-priority missions are delayed in execution when the energy prediction is lower than a preset threshold.

[0020] An autonomous navigation and obstacle avoidance system for a Mars rover, comprising multi-mode switching for performing the multi-mode switching method described above, includes: The central controller is used to run obstacle detection algorithms, ranging algorithms, and obstacle avoidance control algorithms, parse programmable commands, and manage the switching between the three working modes: fully autonomous mode, semi-autonomous mode, and fully programmable mode. The environmental perception module includes a binocular camera and a single-line lidar. The binocular camera is used for environmental perception on the Martian surface and obstacle detection and ranging. The single-line lidar is used for emergency braking safety redundancy in abnormal situations. The autonomous obstacle avoidance module controls the detection vehicle to complete navigation and obstacle avoidance according to the obstacle azimuth angle, distance, width and length parameters fed back by the obstacle ranging module and a preset obstacle avoidance strategy. The wireless communication module is used to receive ground-based programmable commands and transmit image data back to the ground. The tracked chassis drive system includes two DC brushed motors and a metal tracked chassis, which achieve forward, backward and stationary rotation movements through the differential speed of the left and right tracks; The data storage module, including an SD card, is used to locally cache image data to prevent data loss due to communication delays and bandwidth limitations between Mars and Earth.

[0021] Furthermore, the central controller is an NVIDIA Jetson Nano B01 processing board, and the chassis driver board is an STM32F103RCT6 microcontroller, which communicates via a USB bus. The central controller integrates an environmental complexity assessment unit, which calculates the environmental complexity level of the current road conditions based on real-time perception data to support the generation of automatic mode switching suggestions. The system also includes an energy management module, which is used to execute a graded energy-saving strategy and run an energy prediction model. It predicts energy supply trends based on the output current of solar panels, the remaining power of batteries, and historical data on changes in sunlight on the Martian surface, and dynamically allocates the energy budget according to mission priorities.

[0022] Furthermore, the wireless communication module is also used to transmit image data captured by the probe vehicle back to the ground in real time, and to receive image downlink instructions from the ground and upload image data cached locally on the SD card to the ground station; The probe vehicle is also equipped with image storage and image download operations. The image storage operation involves the probe vehicle capturing video images of a preset duration and saving them to an SD card. The image download operation involves the probe vehicle uploading the video data cached in the SD card to the ground station via wireless communication.

[0023] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Flexible switching between three modes. This invention innovatively introduces a semi-autonomous mode and constructs a three-level progressive control architecture of fully autonomous mode, semi-autonomous mode and fully programmable mode, filling the gap between fully autonomous mode and fully programmable mode. In semi-autonomous mode, the probe receives the target direction and travel distance specified by the ground and still autonomously performs obstacle avoidance during travel, taking into account the global decision-making ability of ground experts and the local obstacle avoidance ability of the probe, which is a major improvement over the existing binary mode scheme.

[0024] (2) The safety assurance mechanism is sound. The fully programmable mode forces the shielding of the autonomous obstacle avoidance decision output, which provides a safety guarantee for ground experts to take over the motion control of the exploration vehicle in extreme terrain, effectively preventing the autonomous obstacle avoidance algorithm from misjudging and causing greater danger; at the same time, it is equipped with a single-line lidar as an emergency braking safety redundancy, which together with the shielding mechanism at the control level forms a double layer of safety protection.

[0025] (3) Automatic energy-saving management. All three modes are equipped with an automatic energy-saving mechanism after the mission is completed. After the rover completes the mission instructions for the current stage in any mode, it will automatically perform energy-saving operations such as stopping in place, turning off the camera, and reducing the power consumption of the central controller. In view of the actual constraints of limited energy on Mars, this effectively extends the duration of the exploration mission.

[0026] (4) Communication delay friendly design. Image data is cached in real time through the local SD card caching mechanism to avoid data loss due to link interruption. Combined with the phased task execution mode, the rover can complete all tasks independently after receiving the phased task instructions without real-time communication intervention, effectively solving the problem of communication delay of more than 22 minutes between Mars and Earth.

[0027] (5) The tracked chassis has strong adaptability. The tracked chassis design enables forward, backward and stationary rotation through differential control of the left and right tracks. Compared with the wheeled chassis, it has a larger ground contact area and lower ground contact pressure, making it more suitable for the rugged terrain of Mars and effectively reducing the risk of rollover. Attached Figure Description

[0028] Figure 1 This is an overall flowchart of the multi-mode switching autonomous navigation and obstacle avoidance method for Mars rover according to the present invention; Figure 2 This is a conceptual diagram of the autonomous navigation and obstacle avoidance control system architecture for the Mars rover of this invention. Figure 3(a) is a schematic diagram of establishing the Martian terrain dataset in Step 1 of the training and deployment of the Martian surface obstacle detection model based on YOLOv3 of the present invention; Figure 3(b) is a schematic diagram of network model training data annotation in Step 2 of the training and deployment of the Martian surface obstacle detection model based on YOLOv3 of the present invention; Figure 3(c) is a schematic diagram of YOLO v3 model parameter training in Step 3 of the training and deployment of the Martian surface obstacle detection model based on YOLOv3 of the present invention; Figure 3(d) is a schematic diagram of model deployment in Step 4 of the training and deployment of the Martian surface obstacle detection model based on YOLOv3 of the present invention. Figure 4 This is a schematic diagram illustrating the safety margin calculation and obstacle avoidance in the autonomous obstacle avoidance strategy of this invention; Figure 5 This is a flowchart of the programmable interactive module of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0031] First Embodiment like Figure 1 As shown, this embodiment provides a multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers, including: S1: Acquire environmental perception data around the detection vehicle, and calculate the environmental complexity level of the current road condition based on the environmental perception data.

[0032] In this embodiment, step S1 specifically includes: The environmental perception data is collected by a binocular camera mounted on the detection vehicle. The environmental complexity level is calculated by weighted fusion of feature parameters from three dimensions: obstacle density, terrain undulation, and feasible area ratio. After the probe is powered on, it enters the fully autonomous mode by default. The central controller automatically generates a mode switching suggestion based on the real-time calculated environmental complexity level and sends a confirmation request to the ground through the wireless communication module. The ground completes the final confirmation or rejection of the mode switching by returning a confirmation command or a override command. When communication is interrupted and the environmental complexity level exceeds the emergency threshold, the central controller automatically performs a mode switch and transmits the switch record back to the ground via a low-power beacon.

[0033] In this embodiment, the environmental complexity assessment achieved in step S1 serves as the data foundation for multi-mode switching decisions. Specifically, the construction of the environmental complexity level fully considers the impact of typical Martian surface terrain features on the rover's mobility: obstacle density directly reflects the number of targets the rover needs to avoid within a unit area; terrain undulation characterizes the potential impact of surface slope changes on the rover's driving stability; and the feasible area ratio macroscopically depicts the spatial margin available for safe passage under the current road conditions. By weighted fusion of the above three dimensions of feature parameters, the system can quantify the abstract road condition complexity into a calculable complexity level, providing a reliable quantitative basis for the automatic recommendation of subsequent working modes.

[0034] For example, in one specific embodiment, the obstacle density is calculated as follows: the terrain image captured by the binocular camera is divided into N×N grid cells, and the ratio of the number of grid cells occupied by obstacles (such as rocks, crater edges, etc.) to the total number of grid cells is calculated as the obstacle density feature value of the area. The terrain undulation is calculated as follows: based on the 3D point cloud data reconstructed by binocular stereo vision, the standard deviation of the elevation values ​​in each grid cell is extracted, and normalized using the maximum adaptable slope angle of the vehicle's suspension system as a reference to obtain the terrain undulation feature value. The feasible area proportion is calculated as follows: based on the passability constraints such as the slope angle, roughness, and obstacle height of each grid cell, it is determined whether the grid cell is safe for the vehicle to pass through, and the ratio of the number of feasible grid cells to the total number of grid cells is calculated as the feasible area proportion feature value. The feature parameters of the above three dimensions are assigned preset weight coefficients (such as obstacle density weight 0.4, terrain undulation weight 0.3, and feasible area proportion weight 0.3), and the weighted sum is used to obtain the comprehensive environmental complexity level of the current road condition.

[0035] Regarding the interactive mechanism for mode switching decisions, the rover defaults to fully autonomous mode upon power-up, aiming to minimize reliance on ground communication links under normal road conditions. However, when the level of environmental complexity changes, the central controller does not automatically execute mode switching. Instead, it sends switching suggestions to ground experts for confirmation. This closed-loop decision-making mechanism of "machine recommendation + human confirmation" leverages the rover's real-time perception and rapid assessment capabilities while ensuring the final judgment of ground experts at critical decision-making nodes. The override command grants ground experts the authority to bypass the rover's autonomous decision-making and directly take over control under any circumstances, ensuring the highest priority of human-in-the-loop decision-making.

[0036] To address the unavoidable communication interruption scenarios in Mars exploration missions, this invention also incorporates an autonomous response strategy in step S1. When the communication link is interrupted and the environmental complexity level exceeds a preset emergency threshold, waiting for ground confirmation would cause the rover to remain stranded in a dangerous area for an extended period. In this situation, the central controller will no longer wait for a ground response and will directly execute a mode switch to escape the current danger. Simultaneously, it will intermittently transmit the switching record back to the ground via low-power beacons, allowing the ground to fully trace the rover's autonomous decision-making process after communication is restored. This mechanism ensures the safety of the rover while maintaining the transparency and traceability of the operational process.

[0037] S2: Select the current working mode from the preset fully autonomous mode, semi-autonomous mode and fully programmed mode according to the environmental complexity level, and confirm or override the selected working mode through the mode identifier field of the ground programmed command.

[0038] In this embodiment, the mode selection in step S2 is a key decision-making step following the environmental complexity assessment results of step S1. Specifically, the central controller has a built-in preset mode determination rule, which divides the environmental complexity level calculated in step S1 into several consecutive level intervals, with each level interval corresponding to a recommended working mode. For example, when the environmental complexity level is in the low level interval, it indicates that the current road conditions have sparse obstacles, flat terrain, and ample feasible areas, and the system recommends maintaining or switching to the fully autonomous mode; when the environmental complexity level is in the medium level interval, it indicates that the current road conditions have a certain number of obstacles or terrain undulations that exceed the comfortable processing range of the autonomous algorithm, and the system recommends switching to the semi-autonomous mode, where ground experts specify the macroscopic travel direction and the detection vehicle autonomously completes local obstacle avoidance; when the environmental complexity level is in the high level interval, it indicates that the current road conditions have dense obstacles, severe terrain undulations, or extremely limited feasible areas, and the autonomous obstacle avoidance algorithm has a high risk of misjudgment, and the system recommends switching to the fully programmable mode, where ground experts take complete control. The above recommendations are transmitted to the ground via wireless communication modules in the form of mode switching suggestions for ground operators to refer to and make decisions.

[0039] At the command interaction level, the ground-based programmable commands include a separate mode identifier field. This field contains at least two binary bits, corresponding to three operating states: fully autonomous mode, semi-autonomous mode, and fully programmable mode. When ground operators confirm the system-recommended mode, they return a confirmation command to switch the rover to the corresponding mode. When ground operators believe that the system-recommended mode is not suitable for the current road conditions, they can send a override command carrying a different mode identifier field to directly override the system recommendation and force the rover to switch to the specified operating mode. For example, in certain special scientific exploration scenarios, even if the environmental complexity assessment results show that the current road conditions are suitable for fully autonomous mode, ground experts, for the need for precise positioning of specific high-value exploration targets, can still use an override command to put the rover into semi-autonomous mode, introducing manual path planning guidance while ensuring safety. Similarly, if the environmental complexity assessment in step S1 is misjudged due to a decrease in the confidence level of the binocular camera caused by Martian dust, ground experts can directly correct the system recommendation using an override command to prevent the rover from entering an inappropriate operating mode.

[0040] Through the closed-loop decision-making mechanism of "system recommendation - manual confirmation or override correction" described above, this invention fully utilizes the real-time perception and computing capabilities of the probe vehicle while ensuring the final judgment right of ground experts at key decision-making nodes, thus achieving complementary advantages between autonomous decision-making and manual decision-making.

[0041] S3: Under the confirmed or modified working mode, the vehicle performs the travel task according to the control authority allocation strategy corresponding to the mode, and performs the corresponding level of autonomous obstacle avoidance decision based on obstacle detection and ranging results during the travel. In the fully autonomous mode, the probe vehicle makes completely autonomous decisions and performs navigation and obstacle avoidance. In the semi-autonomous mode, the ground specifies macroscopic target parameters and the probe vehicle autonomously performs local obstacle avoidance. In the fully programmable mode, the ground completely takes over the control and the autonomous obstacle avoidance decision output is forcibly blocked.

[0042] In step S3, the control permission allocation strategy varies depending on the current working mode, specifically as follows: When the current working mode, after confirmation or correction, is the fully autonomous mode, the rover makes decisions entirely autonomously. It uses the obstacle detection module to detect obstacles on the road in real time, and the obstacle ranging module to calculate the three-dimensional coordinates and obstacle avoidance parameters of the obstacles. The autonomous obstacle avoidance module controls the rover to complete navigation and obstacle avoidance according to the obstacle avoidance strategy. After the rover travels a preset distance autonomously, it stops in place, turns off the camera, and enters the energy-saving mode. It continues to travel the preset distance on the next Martian day and then enters the energy-saving mode again. This cycle is repeated and is suitable for conventional exploration routes. The rover relies entirely on its own capabilities for navigation. When the current working mode, after confirmation or correction, is the semi-autonomous mode, the ground specifies the macroscopic task parameters, the probe vehicle autonomously performs local obstacle avoidance, the probe vehicle receives ground program control instructions, autonomously travels a specified distance from the starting point to the target point, performs autonomous obstacle avoidance function during the journey, and stops in place and turns off the camera after the instruction is completed, entering energy-saving mode. This mode is suitable for scenarios where ground experts specify the target direction and travel distance in moderately complex road conditions, and the probe vehicle autonomously completes obstacle avoidance. When the current working mode, after confirmation or correction, is the fully controlled mode, the ground takes complete control and shields the autonomous obstacle avoidance decision output. The probe strictly follows the ground control instructions to rotate the front of the vehicle by a specified angle and move forward a specified distance. During the journey, the obstacle avoidance function is not executed. After the instructions are executed, the probe stops in place and turns off the camera to enter energy-saving mode. This mode is suitable for scenarios where ground experts take complete control in extremely complex environments to prevent misjudgments by the autonomous obstacle avoidance algorithm. In the fully controlled mode, the obstacle avoidance decision output of the autonomous obstacle avoidance module is forcibly shielded during the journey. The execution of the current instructions is only paused when the single-line lidar triggers emergency braking, and execution is resumed after further instructions from the ground.

[0043] In step S3, the autonomous obstacle avoidance decision-making based on the obstacle detection and ranging results is performed at the corresponding level, including the obstacle detection and ranging method, specifically: The autonomous obstacle avoidance decision-making utilizes a deep learning-based target detection algorithm to detect craters, rocks, and steep slopes on the Martian surface in real time. It uses a binocular stereo vision measurement algorithm to calculate the three-dimensional coordinates of the geometric center point of the obstacle and the three-dimensional coordinates of the four vertices of the circumscribed rectangle. Based on the three-dimensional coordinates, it calculates four obstacle avoidance parameters: azimuth, distance, width, and length of the obstacle. Based on the obstacle avoidance parameters and a preset obstacle avoidance strategy, it controls the differential speed between the left and right sides of the tracked chassis to achieve steering and obstacle avoidance on the rugged terrain of Mars.

[0044] The preset obstacle avoidance strategy adopts a four-level progressive obstacle avoidance decision-making mechanism. The obstacle avoidance turning judgment includes a dynamic safety margin, specifically: The four-level progressive obstacle avoidance decision-making mechanism executes differentiated obstacle avoidance responses based on the characteristics of different obstacles on the Martian surface: the first level is the safe passage level, which maintains the original direction of travel when sufficient passage space is detected; the second level is the mild detour level, which calculates the minimum safe turning angle to avoid obstacles when they partially block the path; the third level is the tentative passage level, which approaches the obstacle at a speed lower than the normal travel speed and continuously evaluates it when the obstacle boundary is unclear or the risk is within a preset controllable range, and retreats to a safe position if the risk increases; the fourth level is the return route level, which automatically plans the return route when it is determined that the path ahead is completely impassable. When the distance to an obstacle is less than a preset threshold, obstacle avoidance judgment is initiated. The obstacle is determined to be located to the left, right, or directly in front of the probe based on its azimuth angle. The angle that the front of the probe needs to rotate is calculated based on the width and length of the obstacle and the cross-sectional dimensions of the probe. The probe is then controlled to rotate in place by the left and right tracks at a differential speed to avoid the obstacle. The calculation of the rotation angle includes a dynamic safety margin. The dynamic safety margin is adaptively adjusted according to the current speed of the probe, the friction coefficient of the Martian surface, and the type of obstacle. On typical soft sand or gravel surfaces on Mars, the safety margin is automatically increased to prevent sideslip. The autonomous obstacle avoidance decision-making also includes obstacle avoidance confidence assessment, outputting a confidence score for each obstacle avoidance decision. When the confidence score is lower than a preset low confidence threshold, the probe automatically decelerates to a safe speed and stops moving forward. At the same time, it requests instructions from the ground via the wireless communication module and maintains its current position and a minimum power consumption listening state while waiting for instructions from the ground. When the confidence score is higher than the preset low confidence threshold, the probe maintains the current obstacle avoidance decision and continues to execute.

[0045] Step S3 further includes the detection vehicle being equipped with a binocular camera and a single-line lidar to form a heterogeneous fusion perception system, specifically: The binocular camera is used as the main environmental perception sensor to perform obstacle detection, ranging and terrain modeling on the Martian surface. The single-line lidar is used as an independent safety redundancy channel to trigger emergency braking when the distance between the rover and the obstacle exceeds the preset safety warning line, forming a two-layer protection architecture of binocular camera main perception and lidar safety redundancy. The heterogeneous fusion sensing system is equipped with a cross-validation mechanism for sensing confidence and a redundancy switching mechanism: when the binocular camera's detection confidence drops below a preset cross-validation threshold due to strong changes in lighting on the Martian surface, dust obstruction, or loss of texture features, the system automatically marks the low-confidence area and triggers the measurement data of the single-line lidar to cross-validate the low-confidence area. The system uses the lidar's ranging data to correct the binocular vision's perception results, compensating for blind spots and failure scenarios in visual perception under special lighting conditions on the Martian surface. When the binocular camera experiences a hardware failure or the sensing data remains abnormal, the system degrades to using the ranging data from the single-line lidar to maintain basic emergency braking functionality.

[0046] In this embodiment, step S3 executes differentiated control authority allocation and obstacle avoidance response logic based on the final working mode confirmed or modified in step S2. The division into three modes is essentially a gradual adjustment of the control authority allocation between "ground expert decision-making" and "autonomous vehicle decision-making": the fully autonomous mode grants all motion decision-making authority to the vehicle, which is suitable for conventional detection scenarios with normal communication conditions and low road complexity, and can maximize the autonomous operation efficiency of the vehicle; the semi-autonomous mode retains the macro-path decision-making authority to the ground expert and grants the micro-obstacle avoidance decision-making authority to the vehicle, which is suitable for compromise scenarios in moderately complex road conditions where ground experts are needed to provide global path planning guidance but it is not advisable to rely entirely on remote control. The decision-making authority allocation in this mode reflects the optimal combination between the macro-guidance of human in the loop and the micro-response of machine autonomy; the fully programmable mode returns all motion decision-making authority to the ground expert, which is suitable for high-risk scenarios in extremely complex environments where the autonomous obstacle avoidance algorithm may make misjudgments. In this case, the ground expert performs precise segment-by-segment or action-by-action control of the vehicle based on the transmitted images and telemetry data.

[0047] At the autonomous obstacle avoidance execution level, this invention designs a differentiated processing mechanism based on the physical characteristics of typical obstacles on the Martian surface. A deep learning-based target detection algorithm outputs corresponding category labels and detection bounding boxes for three main types of obstacles on the Martian surface—craters, rocks, and steep slopes. Different types of obstacles trigger different obstacle avoidance strategy branches: craters are recessed obstacles, and their boundaries usually have soft edges, requiring an additional safety margin to prevent edge collapse during obstacle avoidance; rocks are protruding obstacles, and obstacle avoidance is mainly based on the size of their circumscribed rectangle to determine whether they can be bypassed; steep slopes are terrain-related obstacles, and obstacle avoidance requires evaluation based on the rover's maximum climbing angle and longitudinal clearance angle. A binocular stereo vision measurement algorithm calculates the three-dimensional coordinates of the obstacle's geometric center point and the three-dimensional coordinates of the four vertices of its circumscribed rectangle by matching corresponding feature points in the left and right views, thereby deriving four obstacle avoidance parameters relative to the rover: azimuth angle, straight-line distance, lateral width, and longitudinal length. The accuracy of the extraction of the above four parameters directly determines the reliability of subsequent steering and obstacle avoidance decisions. The azimuth angle is used to determine the spatial orientation of the obstacle, the distance is used to assess the urgency of the danger, and the width and length are used to determine whether the probe vehicle can safely bypass or cross the obstacle. Based on this, the present invention adopts a four-level progressive obstacle avoidance decision-making mechanism, judging the passage conditions from low to high according to the obstacle type and spatial distribution characteristics: the first level, the safe passage level, is the optimal result, where the probe vehicle maintains its current direction and passes directly; the second level, the slight detour level, is the most common situation, where the probe vehicle only needs to adjust its course slightly to bypass the obstacle; the third level, the tentative passage level, is suitable for scenarios where the obstacle boundary is unclear or the risk is uncertain, where the probe vehicle approaches tentatively at low speed so that it can retreat in time if the risk increases; the fourth level, the original route return level, is the final backup plan, where the probe vehicle turns around and returns to a safe area to wait for subsequent instructions when the current path is completely blocked. In the aforementioned decision-making process, the safety margin is not a fixed constant, but is dynamically calculated based on the rover's current speed, the friction coefficient of the Martian surface, and the type of obstacle. Under low-adhesion conditions such as soft sand or gravel surfaces, the safety distance is automatically increased, effectively reducing the risk of skidding or getting stuck. The obstacle avoidance confidence assessment quantifies and scores the credibility of each obstacle avoidance decision. When the confidence level is low, the rover actively decelerates and pauses its advance, requesting further decision guidance from the ground. This mechanism builds an additional safety barrier at the boundary of the autonomous obstacle avoidance algorithm's capabilities, preventing the algorithm from blindly executing potentially dangerous obstacle avoidance actions when it is in an "uncertain" state.

[0048] In terms of the perception system architecture, this invention adopts a heterogeneous fusion perception system composed of a binocular camera and a single-line lidar. The two types of sensors complement each other in terms of physical characteristics and information dimensions: the binocular camera can provide rich color and texture information, which is beneficial for deep learning algorithms to accurately classify obstacles and understand semantics. However, it is prone to losing texture features and causing detection failure under conditions of strong light changes on the Martian surface (such as the high contrast between direct midday sunlight and shadow areas) or dust obstruction. The single-line lidar does not depend on ambient lighting conditions and can directly obtain high-precision distance information, but it can only provide distance data on a single scan line and cannot obtain the semantic category of obstacles. The heterogeneous fusion scheme of this invention fully leverages the physical advantages of the two types of sensors—under normal operating conditions, the binocular vision is mainly used for obstacle detection, classification, and terrain modeling, while the lidar is used as an independent monitoring channel to continuously monitor whether there are obstacles in front of the probe that exceed the safety warning line. Once triggered, it independently performs emergency braking independently of the autonomous obstacle avoidance module. This redundancy mechanism, together with the shielding strategy in the fully programmable mode, constitutes a dual safety guarantee of "perception redundancy + decision redundancy". The cross-validation mechanism is automatically activated when the confidence level of binocular vision decreases due to lighting or dust. It uses measured distance data from the lidar to verify and correct the visual perception results, compensating for the inherent limitations of single-vision perception under the harsh lighting conditions on Mars. When the binocular camera experiences a hardware failure, the system can further degrade to a state where the lidar alone maintains basic emergency braking functionality. This ensures that the rover still has a minimum level of safety assurance even in extreme situations where some sensors fail, until ground intervention resolves the fault or a return-to-base command is issued.

[0049] S4: After completing the current stage of the travel task, automatically perform graded energy-saving operations and dynamically adjust the parameters of the next stage of the travel task based on the energy status prediction results.

[0050] In this embodiment, step S4 specifically includes: The graded energy-saving operation includes three levels: Level 1 energy saving, Level 2 energy saving, and Level 3 energy saving. The first level of energy saving is mild energy saving, which includes reducing the sampling frame rate of the binocular camera and reducing the transmission power of the wireless communication module; The second level of energy saving is medium energy saving, which includes turning off the single-line lidar and reducing the frequency of the central processing unit; The third level of energy saving is deep hibernation, which includes shutting down all unnecessary peripherals and switching the central controller to the lowest power standby state, retaining only the timed wake-up and communication monitoring functions; The rover automatically selects the appropriate energy-saving level based on the current mission stage and energy status in different working modes. After completing the mission instructions for the current stage in any mode, the rover automatically performs energy-saving operations: stops in place, turns off the camera, reduces the power consumption of the central controller, and waits for the next round of instructions or the next Martian day. The graded energy-saving operation is also equipped with an energy prediction model. The energy prediction model predicts the energy supply trend in the future time period based on solar panel output current monitoring data, battery remaining power data and historical data on Martian surface illumination changes. Based on the energy supply trend and current mission priority, the energy-saving level and mission execution plan are dynamically adjusted. High-priority scientific exploration missions are given priority in ensuring energy supply, while low-priority missions are delayed in execution when the energy prediction is lower than a preset threshold.

[0051] In this embodiment, the tiered energy-saving operation and energy prediction model implemented in step S4 constitute the energy security system for the rover's long-term autonomous operation on the Martian surface. The design of the three-tiered energy-saving strategy fully considers the energy consumption optimization requirements under different mission intervals. Specifically, Level 1 energy saving (mild energy saving) is suitable for short-term waiting scenarios—for example, when the rover completes a preset distance in fully autonomous mode and waits for the next Martian day to continue operations, reducing the sampling frame rate of the binocular camera can reduce the computational overhead during image data acquisition and transmission, and reducing the transmission power of the wireless communication module can reduce RF power consumption without interrupting communication monitoring; Level 2 energy saving (moderate energy saving) is suitable for medium-to-long-term standby scenarios—turning off the single-line lidar can eliminate the continuous power consumption of this module, and reducing the frequency of the central processing unit can reduce dynamic power consumption and leakage power consumption by lowering the operating voltage and clock frequency; Level 3 energy saving (deep hibernation) is suitable for long-term hibernation scenarios spanning the Martian night—turning off all unnecessary peripherals and switching the central controller to the lowest power standby state, at which time only the timed wake-up function (for automatically restoring to the working state at a preset time) and the communication monitoring function (for receiving emergency instructions that may be sent from the ground) are retained, thereby minimizing the overall power consumption while maintaining basic responsiveness.

[0052] After completing the current stage task instructions in any mode, the probe vehicle uniformly executes energy-saving operations such as "stopping in place, turning off the camera, and reducing the power consumption of the central controller". This universal energy-saving action ensures that the probe vehicle has consistent energy-saving behavior logic when switching between different working modes, and simplifies the state management complexity of the central controller.

[0053] Regarding energy supply forecasting, the Martian surface exhibits distinct diurnal and seasonal light conditions, while Martian dust storms can significantly reduce the output power of solar panels in the short term. The energy forecasting model configured in this invention predicts energy supply trends over the next few hours to days by real-time monitoring of solar panel output current, combined with remaining battery power data and historical data on Martian surface light variations. The forecast results serve as the basis for adjusting energy efficiency levels and mission plans. For example, when the energy forecasting model determines that light conditions will remain favorable and energy supply abundant for a period, the system can prioritize Level 1 energy efficiency to maintain the rapid response capability of sensors, even if the rover is idle. When the forecasting model predicts an impending dust storm or the approaching Martian night, the system switches to Level 3 energy efficiency in advance to maximize the preservation of remaining battery power before the energy shortage period, ensuring the rover does not suffer permanent shutdown due to energy depletion.

[0054] Regarding the dynamic adjustment of mission execution plans, this invention divides scientific exploration missions into two levels: high priority and low priority. High-priority scientific exploration missions are pre-defined key exploration targets with irreplaceable scientific value (such as high-precision imaging of predetermined exploration points, close-range analysis of specific mineral components, etc.). Even under conditions of tight energy supply, the system prioritizes allocating energy budgets for high-priority missions to ensure their smooth execution. Low-priority missions (such as image acquisition of non-critical areas or recording of routine environmental parameters, etc.) are automatically delayed when energy predictions fall below a preset threshold, and will continue when energy supply is restored or high-priority missions are completed. This mission priority-driven energy allocation mechanism ensures that limited electrical energy is always prioritized for exploration activities with the highest scientific value, effectively improving the scientific output per unit of energy consumption, and providing a systematic energy management solution for long-term independent operation of Mars rovers in the extremely energy-constrained deep space environment.

[0055] Second Embodiment This embodiment provides a multi-mode switching autonomous navigation and obstacle avoidance system for a Mars rover to perform the multi-mode switching autonomous navigation and obstacle avoidance method as described in the first embodiment, including: The central controller is used to run obstacle detection algorithms, ranging algorithms, and obstacle avoidance control algorithms, parse programmable commands, and manage the switching between the three working modes: fully autonomous mode, semi-autonomous mode, and fully programmable mode. The environmental perception module includes a binocular camera and a single-line lidar. The binocular camera is used for environmental perception on the Martian surface and obstacle detection and ranging. The single-line lidar is used for emergency braking safety redundancy in abnormal situations. The autonomous obstacle avoidance module controls the detection vehicle to complete navigation and obstacle avoidance according to the obstacle azimuth angle, distance, width and length parameters fed back by the obstacle ranging module and a preset obstacle avoidance strategy. The wireless communication module is used to receive ground-based programmable commands and transmit image data back to the ground. The tracked chassis drive system includes two DC brushed motors and a metal tracked chassis, which achieve forward, backward and stationary rotation movements through the differential speed of the left and right tracks; The data storage module, including an SD card, is used to locally cache image data to prevent data loss due to communication delays and bandwidth limitations between Mars and Earth.

[0056] Furthermore, the central controller is an NVIDIA Jetson Nano B01 processing board, and the chassis driver board is an STM32F103RCT6 microcontroller, which communicates via a USB bus. The central controller integrates an environmental complexity assessment unit, which calculates the environmental complexity level of the current road conditions based on real-time perception data to support the generation of automatic mode switching suggestions. The system also includes an energy management module, which is used to execute a graded energy-saving strategy and run an energy prediction model. It predicts energy supply trends based on the output current of solar panels, the remaining power of batteries, and historical data on changes in sunlight on the Martian surface, and dynamically allocates the energy budget according to mission priorities.

[0057] Furthermore, the wireless communication module is also used to transmit image data captured by the probe vehicle back to the ground in real time, and to receive image downlink instructions from the ground and upload image data cached locally on the SD card to the ground station; The probe vehicle is also equipped with image storage and image download operations. The image storage operation involves the probe vehicle capturing video images of a preset duration and saving them to an SD card. The image download operation involves the probe vehicle uploading the video data cached in the SD card to the ground station via wireless communication.

[0058] Third Embodiment This embodiment further illustrates the present invention based on the accompanying drawings.

[0059] (I) Overall System Architecture like Figure 2As shown, the Mars rover's autonomous navigation and obstacle avoidance control system of the present invention comprises four main functional modules: an obstacle detection module, an obstacle ranging module, an autonomous obstacle avoidance module, and a programmable interaction module. The obstacle detection module uses a deep learning-based target detection algorithm to detect obstacles on the path in real time; the obstacle ranging module uses binocular stereo vision to calculate the three-dimensional coordinates of obstacles and obstacle avoidance parameters; the autonomous obstacle avoidance module controls the rover to complete navigation and obstacle avoidance according to a preset strategy based on obstacle location information; and the programmable interaction module realizes command interaction and image transmission between the ground and the rover through a wireless communication system.

[0060] like Figure 2 As shown, the system hardware is based on a Jetson Nano B01 processing board, connecting a stereo camera and a wireless network card via a USB interface. It communicates with an STM32F103RCT6 chassis driver board via a USB bus. The driver board controls two DC brushed motors to move the metal tracked chassis. A lithium battery provides power to the entire system, while a DC-DC voltage regulator module provides a stable operating voltage for each module.

[0061] like Figure 2As shown, the Mars rover autonomous navigation and obstacle avoidance control system of the present invention uses the Jetson Nano B01 processing board as the core computing unit at the hardware level, and is divided into four major functional modules at the functional level: obstacle detection module, obstacle ranging module, autonomous obstacle avoidance module, and programmable interaction module. The binocular camera connects to the Jetson Nano B01 processing board via a USB interface, enabling power supply and data transmission over a single cable. This allows for the acquisition of binocular images of the Martian surface environment. After receiving the image data, the Jetson Nano B01 processing board runs a deep learning-based target detection algorithm to implement the obstacle detection module function—detecting obstacles such as craters, rocks, and steep slopes in real time. Simultaneously, it runs a binocular stereo vision algorithm to implement the obstacle ranging module function—calculating the obstacle's three-dimensional coordinates and four obstacle avoidance parameters: azimuth, distance, width, and length. The autonomous obstacle avoidance module generates motion control commands based on obstacle location information and preset obstacle avoidance strategies, and sends these commands to the STM32F103RCT6 chassis driver board via the USB bus. The chassis driver board drives two DC brushed motors to operate independently. These two motors drive the tracks on the left and right sides of the metal tracked chassis, respectively. The differential speed movement of the left and right tracks enables the rover to move forward, backward, and rotate in place. The wireless network card also connects to the Jetson Nano B01 processing board via a USB interface, enabling the programmable interaction module to receive ground control commands and transmit image data back to the ground. The lithium battery serves as the vehicle's energy supply unit, providing power to the Jetson Nano B01 processing board, binocular cameras, wireless network card, STM32F103RCT6 chassis drive board, and two DC brushed motors. A DC-DC voltage regulator module converts the lithium battery output voltage into stable operating voltages required by each module, ensuring a stable and reliable power supply for all modules under the complex conditions of the Martian surface.

[0062] (II) Training and Deployment of a YOLOv3-based Obstacle Detection Model for the Martian Surface As shown in Figures 3(a)-3(d), the obstacle detection module of the present invention realizes real-time detection of obstacles on the surface of Mars through the YOLOv3 deep learning network model. Its training and deployment process includes the following four steps.

[0063] Step 1: As shown in Figure 3(a), establish a Martian terrain dataset. First, a Martian topographic image dataset was established. This dataset consists of Martian surface images actually taken by rovers during Mars exploration missions, as well as images of simulated Martian environments collected in ground-based simulated Mars test sites. The images in the dataset must cover typical Martian surface terrain scenes, including but not limited to: flat sandy areas, scree areas, rocky areas, crater edges, gentle slopes, and steep slopes. To ensure the diversity and generalization ability of the dataset, the collected images must include Martian surface scenes under different lighting conditions (light at different solar altitude angles during the Martian day), different shooting angles, and different degrees of dust obstruction.

[0064] Step 2: As shown in Figure 3(b), label the training data for the network model. The acquired Martian surface images were manually annotated, focusing on three typical obstacles on the Martian surface: hills, rocks, and steep slopes. An open-source image annotation tool (such as LabelImg) was used. The annotation method involved marking the location coordinates and category label of each obstacle in the image with a rectangular bounding box, generating an annotation file conforming to the YOLOv3 training format. The dataset was then divided into training, validation, and test sets according to a preset ratio.

[0065] Step 3: As shown in Figure 3(c), YOLOv3 network model parameter training. First, a YOLOv3 network model parameter training environment was set up, including installing the Ubuntu operating system, CUDA parallel computing architecture, cuDNN deep neural network acceleration library, and either Darknet or PyTorch deep learning framework. The input image size was uniformly scaled to a preset size (e.g., 416×416 pixels). The YOLOv3 network used Darknet-53 as the feature extraction backbone, extracting multi-level semantic features of obstacles in the Martian surface image through residual structure. The network output layer used three detection layers of different scales to detect obstacles of different sizes, achieving accurate recognition of multi-scale obstacles on the Martian surface. Network training hyperparameters (including batch size, learning rate, momentum parameter, and weight decay coefficient) were set. The YOLOv3 network was iteratively trained using the training set. The CIoU loss function was used to calculate the localization loss, confidence loss, and classification loss between the predicted bounding boxes and the ground truth bounding boxes. The network weight parameters were updated using the backpropagation algorithm. After each preset training round, the current model is validated and evaluated using a validation set. Evaluation metrics such as mean precision (mAP) and recall are calculated. Training is terminated early when the validation set loss value no longer decreases for several consecutive rounds. After training is complete, the final model is comprehensively evaluated using a test set to verify its detection accuracy and generalization ability in the complex environment of the Martian surface.

[0066] Step 4: As shown in Figure 3(d), deploy the model. After training, the trained YOLOv3 weight file and configuration file are ported to the onboard central controller of the probe vehicle—in this embodiment, this is... Figure 2 The Jetson Nano B01 processing board is shown. Before deployment, an inference runtime environment needs to be set up on the Jetson Nano B01 processing board: install the JetPack SDK (including CUDA for ARM, cuDNN, and TensorRT acceleration libraries) and configure the OpenCV image processing library. TensorRT is used to optimize and quantize the trained YOLOv3 model, quantizing the 32-bit floating-point model parameters into 16-bit floating-point numbers or 8-bit integers to reduce the model's memory usage and inference latency on edge devices. After deployment, when the rover performs its mission, the real-time Martian surface images captured by the binocular cameras are directly input to the YOLOv3 detection model deployed on the Jetson Nano B01 processing board. The model outputs the category labels and bounding box location information of each obstacle in the image, providing spatial location and category information of obstacles for the subsequent obstacle ranging module and autonomous obstacle avoidance module.

[0067] (III) Autonomous Obstacle Avoidance Strategy like Figure 4 As shown, the autonomous obstacle avoidance strategy designed in this invention makes decisions based on four parameters: obstacle azimuth angle θ, distance S, width W, and length L. The obstacle avoidance process is as follows: (1) When the distance between the obstacle and the detection vehicle is less than a preset threshold (e.g., 5 meters), the system initiates obstacle avoidance judgment; (2) Determine whether the obstacle is located to the left, right or directly in front of the detection vehicle based on the obstacle's azimuth angle θ; (3) Based on the width W and length L of the obstacle, and combined with the dimensions of the outer rectangle of the cross-section of the probe vehicle (width W_c, height H_c), calculate the spatial relationship between the minimum cross-section for safe passage of the probe vehicle and the outer rectangle of the obstacle; (4) If there is a safe passage space in the current direction, continue to move forward; otherwise, calculate the angle θ_rotate that the front of the vehicle needs to rotate, control the probe vehicle to rotate in place by the differential speed of the left and right tracks by the angle θ_rotate, avoid the obstacle and continue to move forward; (5) The calculation of the rotation angle θ_rotate takes into account the safety margin to ensure that the probe vehicle maintains sufficient lateral distance from the obstacle and avoids collision.

[0068] (iv) Programmable Interaction Module like Figure 5 As shown, the workflow of the program-controlled interactive module is as follows: (1) The wireless communication module receives ground program control commands and forwards the commands to the central controller; (2) The central controller parses the mode identifier field and business logic field in the program control instructions; (3) Based on the mode identifier, perform five operations respectively: fully autonomous mode, semi-autonomous mode, fully program-controlled mode, image storage or image download; (4) Fully autonomous mode: After the rover moves forward 3 meters, it waits, enters energy-saving mode, and then repeats the operation for the second Martian day; (5) Semi-autonomous mode: The probe moves a certain distance according to the programmed instructions, performs autonomous obstacle avoidance during the journey, and enters energy-saving mode after reaching the target point; (6) Fully programmable mode: The probe rotates its front end at a certain angle according to the programmable command, then moves forward a certain distance without performing obstacle avoidance. After reaching the target point, it enters energy-saving mode. (7) Image storage: The probe takes 10 seconds of video images and saves them to the SD card, overwriting the previous data each time it is saved; (8) Image downlink: The probe vehicle uploads the video data stored in the SD card to the ground via wireless communication.

[0069] (V) Semi-autonomous mode experimental verification To verify the effectiveness of the semi-autonomous mode, simulated Martian obstacles (irregularly shaped foam props painted in a tan color) were randomly and densely arranged evenly in the experimental area. After powering on the rover, a semi-autonomous mode programming command was sent via Wi-Fi: instructing the rover to rotate its front 15 degrees clockwise and then move 1 meter. The experimental results showed that the rover successfully adjusted its direction according to the command, autonomously identified and avoided all obstacles during its movement, had sufficient clearance between its cross-section and the obstacles, and there was no risk of collision. After reaching the target point, it automatically stopped and entered energy-saving mode.

[0070] (vi) Experimental verification of fully programmable mode To verify the effectiveness of the fully controlled mode, a fully controlled mode command was sent at the same experimental site: the probe was instructed to rotate its front 15 degrees counterclockwise and then move forward 1 meter, without executing obstacle avoidance. The experimental results showed that the probe strictly followed the command to perform the rotation and forward movement, without triggering the autonomous obstacle avoidance logic, thus verifying that the autonomous obstacle avoidance function was correctly disabled in the fully controlled mode.

[0071] A computer-readable storage medium stores computer code that, when executed, performs the methods described above. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0072] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0073] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0074] It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-mode switching autonomous navigation and obstacle avoidance method for a Mars rover, characterized in that, include: S1: Acquire environmental perception data around the detection vehicle, and calculate the environmental complexity level of the current road condition based on the environmental perception data; S2: Select the current working mode from the preset fully autonomous mode, semi-autonomous mode and fully programmed mode according to the environmental complexity level, and confirm or override the selected working mode through the mode identifier field of the ground programmed command. S3: In the confirmed or modified working mode, the vehicle performs the travel task according to the control authority allocation strategy corresponding to the mode, and performs the corresponding level of autonomous obstacle avoidance decision based on obstacle detection and ranging results during the travel. In the fully autonomous mode, the probe vehicle makes completely autonomous decisions and performs navigation and obstacle avoidance. In the semi-autonomous mode, the ground specifies macroscopic target parameters and the probe vehicle autonomously performs local obstacle avoidance. In the fully programmable mode, the ground completely takes over the control and the autonomous obstacle avoidance decision output is forcibly blocked. S4: After completing the current stage of the travel task, automatically perform graded energy-saving operations and dynamically adjust the parameters of the next stage of the travel task based on the energy status prediction results.

2. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 1, characterized in that, In step S1, environmental perception data around the detection vehicle is acquired, and the environmental complexity level of the current road condition is calculated based on the environmental perception data. Specifically: The environmental perception data is collected by a binocular camera mounted on the detection vehicle. The environmental complexity level is calculated by weighted fusion of feature parameters from three dimensions: obstacle density, terrain undulation, and feasible area ratio. After the probe is powered on, it enters the fully autonomous mode by default. The central controller automatically generates a mode switching suggestion based on the real-time calculated environmental complexity level and sends a confirmation request to the ground through the wireless communication module. The ground completes the final confirmation or rejection of the mode switching by returning a confirmation command or a override command. When communication is interrupted and the environmental complexity level exceeds the emergency threshold, the central controller automatically performs a mode switch and transmits the switch record back to the ground via a low-power beacon.

3. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 1, characterized in that, In step S3, the control permission allocation strategy varies depending on the current working mode, specifically as follows: When the current working mode, after confirmation or correction, is the fully autonomous mode, the rover makes decisions entirely autonomously. It uses the obstacle detection module to detect obstacles on the road in real time, and the obstacle ranging module to calculate the three-dimensional coordinates and obstacle avoidance parameters of the obstacles. The autonomous obstacle avoidance module controls the rover to complete navigation and obstacle avoidance according to the obstacle avoidance strategy. After the rover travels a preset distance autonomously, it stops in place, turns off the camera, and enters the energy-saving mode. It continues to travel the preset distance on the next Martian day and then enters the energy-saving mode again. This cycle is repeated and is suitable for conventional exploration routes. The rover relies entirely on its own capabilities for navigation. When the current working mode, after confirmation or correction, is the semi-autonomous mode, the ground specifies the macroscopic task parameters, the probe vehicle autonomously performs local obstacle avoidance, the probe vehicle receives ground program control instructions, autonomously travels a specified distance from the starting point to the target point, performs autonomous obstacle avoidance function during the journey, and stops in place and turns off the camera after the instruction is completed, entering energy-saving mode. This mode is suitable for scenarios where ground experts specify the target direction and travel distance in moderately complex road conditions, and the probe vehicle autonomously completes obstacle avoidance. When the current working mode, after confirmation or correction, is the fully controlled mode, the ground takes complete control and shields the autonomous obstacle avoidance decision output. The probe strictly follows the ground control instructions to rotate the front of the vehicle by a specified angle and move forward a specified distance. During the journey, the obstacle avoidance function is not executed. After the instructions are executed, the probe stops in place and turns off the camera to enter energy-saving mode. This mode is suitable for scenarios where ground experts take complete control in extremely complex environments to prevent misjudgments by the autonomous obstacle avoidance algorithm. In the fully controlled mode, the obstacle avoidance decision output of the autonomous obstacle avoidance module is forcibly shielded during the journey. The execution of the current instructions is only paused when the single-line lidar triggers emergency braking, and execution is resumed after further instructions from the ground.

4. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 1, characterized in that, In step S3, the autonomous obstacle avoidance decision-making based on the obstacle detection and ranging results is performed at the corresponding level, including the obstacle detection and ranging method, specifically: The autonomous obstacle avoidance decision-making utilizes a deep learning-based target detection algorithm to detect craters, rocks, and steep slopes on the Martian surface in real time. It uses a binocular stereo vision measurement algorithm to calculate the three-dimensional coordinates of the geometric center point of the obstacle and the three-dimensional coordinates of the four vertices of the circumscribed rectangle. Based on the three-dimensional coordinates, it calculates four obstacle avoidance parameters: azimuth, distance, width, and length of the obstacle. Based on the obstacle avoidance parameters and a preset obstacle avoidance strategy, it controls the differential speed between the left and right sides of the tracked chassis to achieve steering and obstacle avoidance on the rugged terrain of Mars.

5. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 4, characterized in that, The preset obstacle avoidance strategy adopts a four-level progressive obstacle avoidance decision-making mechanism. The obstacle avoidance turning judgment includes a dynamic safety margin, specifically: The four-level progressive obstacle avoidance decision-making mechanism executes differentiated obstacle avoidance responses based on the characteristics of different obstacles on the Martian surface: the first level is the safe passage level, which maintains the original direction of travel when sufficient passage space is detected; the second level is the mild detour level, which calculates the minimum safe turning angle to avoid obstacles when they partially block the path; the third level is the tentative passage level, which approaches the obstacle at a speed lower than the normal travel speed and continuously evaluates it when the obstacle boundary is unclear or the risk is within a preset controllable range, and retreats to a safe position if the risk increases; the fourth level is the return route level, which automatically plans the return route when it is determined that the path ahead is completely impassable. When the distance to an obstacle is less than a preset threshold, obstacle avoidance judgment is initiated. The obstacle is determined to be located to the left, right, or directly in front of the probe based on the obstacle's azimuth angle. The angle that the front of the probe needs to rotate is calculated based on the obstacle's width and length combined with the probe's cross-sectional dimensions. The probe is then controlled to rotate in place by the angle through the differential speed of its left and right tracks to avoid the obstacle. The calculation of the rotation angle includes a dynamic safety margin. The dynamic safety margin is adaptively adjusted based on the probe's current speed, the friction coefficient of the Martian surface, and the type of obstacle. On typical soft sand or gravel surfaces on Mars, the safety margin is automatically increased to prevent sideslip. The autonomous obstacle avoidance decision-making also includes obstacle avoidance confidence assessment, outputting a confidence score for each obstacle avoidance decision. When the confidence score is lower than a preset low confidence threshold, the probe automatically decelerates to a safe speed and stops moving forward. At the same time, it requests instructions from the ground via the wireless communication module and maintains its current position and a minimum power consumption listening state while waiting for instructions from the ground. When the confidence score is higher than the preset low confidence threshold, the probe maintains the current obstacle avoidance decision and continues to execute.

6. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 1, characterized in that, Step S3 further includes the detection vehicle being equipped with a binocular camera and a single-line lidar to form a heterogeneous fusion perception system, specifically: The binocular camera is used as the main environmental perception sensor to perform obstacle detection, ranging and terrain modeling on the Martian surface. The single-line lidar is used as an independent safety redundancy channel to trigger emergency braking when the distance between the rover and the obstacle exceeds the preset safety warning line, forming a two-layer protection architecture of binocular camera main perception and lidar safety redundancy. The heterogeneous fusion sensing system is equipped with a cross-validation mechanism for sensing confidence and a redundancy switching mechanism: when the binocular camera's detection confidence drops below a preset cross-validation threshold due to strong changes in lighting on the Martian surface, dust obstruction, or loss of texture features, the system automatically marks the low-confidence area and triggers the measurement data of the single-line lidar to cross-validate the low-confidence area. The system uses the lidar's ranging data to correct the binocular vision's perception results, compensating for blind spots and failure scenarios in visual perception under special lighting conditions on the Martian surface. When the binocular camera experiences a hardware failure or the sensing data remains abnormal, the system degrades to using the ranging data from the single-line lidar to maintain basic emergency braking functionality.

7. The multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers according to claim 1, characterized in that, In step S4, the automatic execution of graded energy-saving operations and the dynamic adjustment of the travel task parameters for the next stage based on the energy state prediction results are specifically as follows: The graded energy-saving operation includes three levels: Level 1 energy saving, Level 2 energy saving, and Level 3 energy saving. The first level of energy saving is mild energy saving, which includes reducing the sampling frame rate of the binocular camera and reducing the transmission power of the wireless communication module; The second level of energy saving is medium energy saving, which includes turning off the single-line lidar and reducing the frequency of the central processing unit; The third level of energy saving is deep hibernation, which includes shutting down all unnecessary peripherals and switching the central controller to the lowest power standby state, retaining only the timed wake-up and communication monitoring functions; The rover automatically selects the appropriate energy-saving level based on the current mission stage and energy status in different working modes. After completing the mission instructions for the current stage in any mode, the rover automatically performs energy-saving operations: stops in place, turns off the camera, reduces the power consumption of the central controller, and waits for the next round of instructions or the next Martian day. The graded energy-saving operation is also equipped with an energy prediction model. The energy prediction model predicts the energy supply trend in the future time period based on solar panel output current monitoring data, battery remaining power data and historical data on Martian surface illumination changes. Based on the energy supply trend and current mission priority, the energy-saving level and mission execution plan are dynamically adjusted. High-priority scientific exploration missions are given priority in ensuring energy supply, while low-priority missions are delayed in execution when the energy prediction is lower than a preset threshold.

8. A Mars rover autonomous navigation and obstacle avoidance system for performing the multi-mode switching autonomous navigation and obstacle avoidance method for Mars rovers as described in any one of claims 1-7, characterized in that, include: The central controller is used to run obstacle detection algorithms, ranging algorithms, and obstacle avoidance control algorithms, parse programmable commands, and manage the switching between the three working modes: fully autonomous mode, semi-autonomous mode, and fully programmable mode. The environmental perception module includes a binocular camera and a single-line lidar. The binocular camera is used for environmental perception on the Martian surface and obstacle detection and ranging. The single-line lidar is used for emergency braking safety redundancy in abnormal situations. The autonomous obstacle avoidance module controls the detection vehicle to complete navigation and obstacle avoidance according to the obstacle azimuth angle, distance, width and length parameters fed back by the obstacle ranging module and a preset obstacle avoidance strategy. The wireless communication module is used to receive ground-based programmable commands and transmit image data back to the ground. The tracked chassis drive system includes two DC brushed motors and a metal tracked chassis, which achieve forward, backward and stationary rotation movements through the differential speed of the left and right tracks; The data storage module, including an SD card, is used to locally cache image data to prevent data loss due to communication delays and bandwidth limitations between Mars and Earth.

9. The multi-mode switching Mars rover autonomous navigation and obstacle avoidance system according to claim 8, characterized in that, The central controller is an NVIDIA Jetson Nano B01 processing board, and the chassis driver board is an STM32F103RCT6 microcontroller. The two communicate via a USB bus. The central controller integrates an environmental complexity assessment unit, which calculates the environmental complexity level of the current road conditions based on real-time perception data to support the generation of automatic mode switching suggestions. The system also includes an energy management module, which is used to execute a graded energy-saving strategy and run an energy prediction model. It predicts energy supply trends based on the output current of solar panels, the remaining power of batteries, and historical data on changes in sunlight on the Martian surface, and dynamically allocates the energy budget according to mission priorities.

10. The multi-mode switching Mars rover autonomous navigation and obstacle avoidance system according to claim 8, characterized in that, The wireless communication module is also used to transmit image data captured by the probe vehicle back to the ground in real time, and to receive image downlink instructions from the ground and upload image data cached locally on the SD card to the ground station. The probe vehicle is also equipped with image storage and image download operations. The image storage operation involves the probe vehicle capturing video images of a preset duration and saving them to an SD card. The image download operation involves the probe vehicle uploading the video data cached in the SD card to the ground station via wireless communication.

Images