Autonomous driving system of robot in which controllable local avoidance is implemented

Abstract

An autonomous driving system of a robot in which controllable local avoidance is implemented is disclosed. The autonomous driving system of a robot is a computing unit for implementing global path planning for generating a global path from a starting position of the robot to a goal position of the robot, and local path planning for generating, around the current position of the robot, a local path that is narrower than the global path, wherein, in the local path planning, the computing unit sets a group of lane paths including the global path and extending in parallel with the global path along the width dimension or the lane dimension of the global path, computes a cost function for each of the set group of lane paths, and generates, as a local path to be followed by the robot, a lane path alternatively selected, among the group of lane paths, from the computed cost functions.

Description

Autonomous driving system for a robot with controllable local avoidance

[0001] The present invention relates to an autonomous driving system for a robot, and more particularly, to an autonomous driving system for a robot in which controllable local avoidance is implemented.

[0002] Robots that transport logistics, such as cargo or products of various sizes in warehouses and production lines, can be operated in autonomous driving mode, where they recognize their current location based on an autonomous driving algorithm, generate an optimized path to the target location, and are controlled to follow that path.

[0003] In autonomous navigation of such robots, path planning, which plans the path of the robot from the starting position of the robot or the loading position for loading the transport object to the goal position of the robot or the unloading position for unloading the transport object, or motion planning, which plans the pose of the robot including the heading or rotation of the robot along the path of the robot, together with the position of the robot as the state of the robot at each position of the robot, can be planned, global path planning, which plans the optimal global path from the starting position of the robot or the starting pose of the robot to the goal position of the robot or the goal pose of the robot, and local path planning for obstacle avoidance or collision avoidance maneuver according to observation observed in real time from the sensors of the robot can be planned, and in addition to the global path planning and local path planning as described above, behavior for defining the interaction with objects in the surrounding environment surrounding the robot can be planned. You can also plan your behavior.

[0004] One embodiment of the present invention includes an autonomous driving system for a robot in which controllable local avoidance is implemented.

[0005] In order to solve the above and other problems, the autonomous driving system of the robot with controllable local avoidance of the present invention is

[0006] A computation unit for implementing global path planning for generating a global path from the robot's starting position to the robot's target position, and local path planning for generating a local path narrower than the global path centered on the robot's current position.

[0007] The above operation unit is,

[0008] In the above local path planning, a group of lane paths are set that include the global path and extend parallel to the global path along the width dimension or lane dimension of the global path,

[0009] A cost function is calculated for each of a set of lane paths, and a lane path selected from the set of lane paths based on the calculated cost function is generated as a local path to be followed by the robot.

[0010] In one embodiment of the present invention, by calculating a candidate for a local path based on a rule set in advance, and selecting or generating a local path for a robot to follow according to a cost function among the candidates for the local paths calculated in this way, it is possible to prevent in advance the occurrence of a safety accident due to an excessive increase in the path cost predicted in the original global path planning due to an excessive obstacle avoidance or excessive collision avoidance maneuver that excessively deviates from the global path generated through global path planning, or an excessive obstacle avoidance or excessive collision avoidance maneuver that deviates from the prediction.

[0011] In one embodiment of the present invention, by setting candidates for local passes in a rule-based manner, the degree of freedom (DOF) for generating local passes can be limited, and local passes can be generated within a controllable prediction range or a predictable range, thereby suppressing the occurrence of safety accidents due to an increase in path cost that excessively deviates from the initial prediction and excessive obstacle avoidance or collision avoidance maneuvers that deviate from the prediction.

[0012] In one embodiment of the present invention, in calculating a cost function for each candidate of a plurality of rule-based local paths, a penalty is set to differentially provide a bias toward a tendency to move away from an obstacle, a bias toward a global path, and a bias toward a current position, thereby inducing the generation of a local path adjacent to the global path planned in the original global path planning or a local path adjacent to the current position of the robot as much as possible within a range capable of sufficiently avoiding obstacles, thereby suppressing an increase in the initially predicted path cost from an excessive avoidance maneuver that excessively circumvents the global path planned in the original global path planning or the current position of the robot, and preventing a safety accident from a predictable avoidance maneuver.

[0013] Figure 1 illustrates a diagram for explaining global path planning.

[0014] FIG. 2 is a drawing for explaining local path planning, which is a drawing for explaining local path planning that sets a local path that bypasses the global path generated from the global path planning illustrated in FIG. 1 for obstacle avoidance or collision avoidance maneuvers.

[0015] FIG. 3 illustrates a diagram for explaining local path planning, which includes a global path as a candidate group of local paths set based on a preset rule, a group of lane paths set on both sides of a global path, a driving section including a local path selected from a cost function for each of the set group of lane paths, and a lane transition section including a bridge path that is incorporated as part of the local path to connect a pose of a robot on the selected local path and a current pose of the robot.

[0016] FIG. 4 illustrates a diagram for explaining a driving section and a lane transition section for a local pass generated from local pass planning, a local pass for avoiding static obstacles and dynamic obstacles, and a first type penalty set as a Gaussian distribution that gradually decreases along the obstacle and its periphery to provide a bias in the tendency to move away from the obstacle.

[0017] Figure 5 illustrates a diagram illustrating a first type penalty, which is set to a Gaussian distribution with a peak section extending as wide as the obstacle and gradually decreasing as it moves away from the peak section, to provide a bias toward moving away from the obstacle.

[0018] Figures 6a and 6b illustrate diagrams for explaining a second type penalty that induces the creation of a local pass as close to the global pass as possible, according to a second type penalty that tends to lead to a global pass within a range that can sufficiently avoid obstacles on the global pass.

[0019] FIGS. 7a to 7c illustrate a method for setting a local path outside the global path within a range sufficient to avoid an obstacle from the global path according to a first type penalty that provides a bias toward a tendency to move away from an obstacle (see FIG. 7a), and a method for setting a local path that returns to the global path again according to a second type penalty that provides a bias toward a tendency to move toward the global path from the position of the robot that has left the obstacle (see FIG. 7b). Different drawings are shown to illustrate the setting of local paths for each section that unfolds according to the robot's movement.

[0020] FIG. 8 illustrates a diagram for explaining a fourth type penalty for setting a local pass from a lane pass of a preferred side position among one side position and the other side position according to a fourth type penalty that is set unevenly for a group of lane passes of one side position and a group of lane passes of the other side position based on a global pass among a group of candidates for a local pass set on both sides of a global pass.

[0021] Figure 9 illustrates a table and diagram for explaining selecting a local pass from a lane pass (e.g., right side avoidance) of one side or the other side based on the global pass (Type 4 penalty) within a range where an obstacle can be sufficiently avoided (Type 1 penalty) or as close as possible to the global pass (Type 2 penalty) according to a cost function that synthesizes Type 1 to Type 4 penalties.

[0022] In order to solve the above and other problems, the autonomous driving system of the robot with controllable local avoidance of the present invention is

[0023] A computation unit for implementing global path planning for generating a global path from the robot's starting position to the robot's target position, and local path planning for generating a local path narrower than the global path centered on the robot's current position.

[0024] The above operation unit is,

[0025] In the above local path planning, a group of lane paths are set that include the global path and extend parallel to the global path along the width dimension or lane dimension of the global path,

[0026] A cost function is calculated for each of a set of lane paths, and a lane path selected from the set of lane paths based on the calculated cost function is generated as a local path to be followed by the robot.

[0027] For example, the above operation unit,

[0028] Rule-based setting of candidates for local passes by forming a group of lane passes as candidates for local passes along the pre-set lane spacing on both sides of the above global pass,

[0029] A lane path to be followed by the robot can be selected from the candidates of the above-described area paths.

[0030] For example, the above operation unit,

[0031] By setting the above group of lane paths symmetrically on both sides of the above global pass,

[0032] The above global pass may be a 0th lane pass, a pair of 1st lane passes may be set symmetrically on both outer sides of the 0th lane pass, and a pair of 2nd lane passes may be set symmetrically on both outer sides of the pair of 1st lane passes.

[0033] For example, the above operation unit,

[0034] A local path can be generated to avoid obstacles recognized from the current location of the robot or to follow a global path from the current location of the robot.

[0035] For example, the above cost function is,

[0036] 1) A first type penalty to provide a bias towards moving away from the perceived obstacle from the robot's current position;

[0037] 2) a second type penalty to provide a bias towards global passes; and

[0038] 3) It can be derived from a penalty that provides at least one bias among the third type penalties for providing a bias in the tendency to move toward the current position of the robot.

[0039] For example, the above cost function is,

[0040] It can be calculated from the sum of the first type penalty, the second type penalty, and the third type penalty.

[0041] For example, the weights of the first type penalty, the second type penalty and the third type penalty are,

[0042] It can be awarded in the following order: Type 1 penalty > Type 2 penalty > Type 3 penalty.

[0043] For example, depending on the magnitude of the weights for the first type penalty, the second type penalty and the third type penalty,

[0044] The above operation unit is,

[0045] Generate a global path and an adjacent local path within a sufficient range for obstacle avoidance or collision avoidance, or

[0046] A local path can be generated from the robot's current position toward the global path within a sufficient range for obstacle avoidance or collision avoidance.

[0047] For example, the above operation unit,

[0048] For the second type penalty and the third type penalty of the second type penalty for providing a bias toward the global pass in the above 2) and the third type penalty for providing a bias toward the current position of the robot in the above 3), respectively,

[0049] According to the relationship between the second type unit scale and the third type unit scale given to each unit transition corresponding to the unit lane pass,

[0050] A local path can be generated from the current position of the robot outside the above global path toward the above global path.

[0051] For example, the cost function is calculated for each of a group of lane passes set on both sides of the global pass, including the global pass,

[0052] A fourth type penalty may be further included to suppress the creation of a local minimum in the selection of a lane pass so that the cost function is minimized, while being applied disproportionately to lane passes at one side and lane passes at the other side centered on the above global pass.

[0053] For example, a Type 4 penalty may not be equally applied to lane passes at one side of the global pass, but a Type 4 penalty may be equally applied to lane passes at the other side of the global pass, but the weight of the Type 4 penalty may be lower than each of the Type 1 penalty, the Type 2 penalty, and the Type 3 penalty.

[0054] For example, the above operation unit,

[0055] A local path toward the global path can be generated from a starting position where autonomous driving of the robot starts or an estimated position with respect to the starting position.

[0056] For example, the above operation unit,

[0057] For the above group of lane passes as candidates for the above local pass, a first type penalty can be added that includes a peak section corresponding to the width of the obstacle and decreases in a Gaussian distribution as it goes beyond the width of the obstacle or the peak section.

[0058] For example, the above operation unit,

[0059] As a candidate for the above-mentioned local pass, for the above-mentioned group of lane passes, a second penalty of a second-type unit scale can be added for each unit transition corresponding to a unit lane path from the above-mentioned global pass.

[0060] For example, the above operation unit,

[0061] As a candidate for the above-mentioned area pass, for the above-mentioned group of lane passes, a third-class penalty of a third-class unit scale may be added for each unit transition corresponding to a unit lane path from the current position of the robot or the pass lane closest to the current position of the robot.

[0062] For example, the above operation unit,

[0063] Generate a bridge path to connect the current location of the robot and a lane path selected from a cost function among the group of lane paths as candidates for the local path,

[0064] The above-generated bridge pass can be incorporated into the local pass together with the lane pass selected from the cost function.

[0065] For example, the above operation unit,

[0066] From the above local path planning, a local path can be generated that includes a lane path selected from the cost function and a bridge path connected to the lane path selected from the cost function.

[0067] For example, the above operation unit,

[0068] Motion planning can be implemented to plan the pose of the robot according to each position along the bridge path so as to connect the pose of the robot determined in advance with respect to the current position of the robot forming the two end positions of the bridge path and the pose of the robot planned in advance on the lane path selected from the cost function.

[0069] For example, the pose of the robot on the bridge path planned from the above local path planning may include the position of the robot and the heading or rotation of the robot.

[0070] For example, the above operation unit,

[0071] Generate a bridge path connecting the current pose of the robot forming the two end positions of the bridge path and the pose of the robot on the lane path selected from the cost function,

[0072] A bridge path can be created that smoothly connects the poses of the robots at the two end positions so that no discontinuity or singularity is formed along the bridge path.

[0073] For example, the above operation unit,

[0074] It is possible to create a bridge path that does not include discontinuous curvature along the above bridge path.

[0075] For example, the above operation unit,

[0076] A bridge path can be generated that includes curved segments that include continuous curvature along the bridge path.

[0077] For example, the above operation unit,

[0078] To avoid including discontinuous curvature, the bridge path is generated in the form of a polynomial of degree 3 or higher, or

[0079] Generate the bridge path from a heuristic estimate to avoid discontinuous curvature, or

[0080] The above bridge path can be generated by smoothing the discontinuous curvature.

[0081] For example, the above operation unit,

[0082] Generate a bridge path connecting the current pose of the robot forming the two end positions of the bridge path and the pose of the robot on the lane path selected from the cost function,

[0083] A bridge path of extended or shortened length can be generated in conjunction with the driving speed of the robot.

[0084] Hereinafter, with reference to the attached drawings, an autonomous driving system of a robot implementing controllable local avoidance according to one embodiment of the present invention will be described.

[0085] Figure 1 illustrates a diagram for explaining global path planning.

[0086] FIG. 2 is a drawing for explaining local path planning, which is a drawing for explaining local path planning that sets a local path that bypasses the global path generated from the global path planning illustrated in FIG. 1 for obstacle avoidance or collision avoidance maneuvers.

[0087] In one embodiment of the present invention, path planning for generating a feasible path between different positions of the robot or different poses of the robot including positions and heading or rotation of the robot within an environment surrounding the robot may include global path planning for generating a global path along the entire driving path of the robot from a start position or a start pose of the robot to a target position or a target force of the robot, and local path planning for generating a local path for a subset of the entire driving path.

[0088] For example, in one embodiment of the present invention, the path planning may include global path planning for generating an optimal global path from a start position or a start pose to a target position or a destination pose based on spatial information about the surrounding environment surrounding the robot, behavior planning for defining an interaction with an object (e.g., an obstacle) in the surrounding environment surrounding the robot, and local path planning for generating a local path that can sufficiently avoid an obstacle or collision as an obstacle avoidance or collision avoidance maneuver. For example, in the path planning according to one embodiment of the present invention, global path planning, behavior planning, and local path planning may be planned in this order.

[0089] Throughout this specification, path planning may mean path planning in a narrow sense that plans a path from a start position of a robot to a target position of the robot, or may be used in a broad sense that includes motion planning that plans a pose of the robot, including, for example, heading or rotation of the robot, as a state of the robot at each position of the robot along a path from the start position of the robot to the target position of the robot, together with the position of the robot at each position of the robot. In the broad sense of path planning that includes motion planning, the state of the robot may include information such as the driving speed or driving acceleration of the robot at each position from the start position of the robot to the target position.

[0090] In one embodiment of the present invention, in the global path planning, an optimal global path can be generated from a starting position of the robot or a starting pose of the robot to a goal position of the robot or a goal pose of the robot. For example, in one embodiment of the present invention, the optimal global path generated from the global path planning may mean the shortest distance (for example, the path cost is minimized) from the starting position of the robot to the goal position of the robot, and the global path planning can generate the shortest path between the starting position and the goal position of the robot as an optimal global path from an algorithm such as a search-based A star search (or A star planner) and a sampling-based RRT (rapidly exploring random tree).

[0091] For example, in global path planning according to one embodiment of the present invention, an A star search may be applied as a search algorithm that searches for a path by using a graph composed of nodes and lines connecting each node from the start position of the robot to the target position, or by using a tree in which each node has at most one parent node, a root node without a parent, and a leaf node without a child. For example, in the A star search, an A star score is calculated for each node, and an optimal path or a shortest path with a minimized cost can be searched for by considering the entire path from the start position to the target position through the A star score that is the sum of the cost of the path from the start position to each node and a heuristic (e.g., an estimated value of the cost of the path) from each node to the target position, and an optimal path or a shortest path with a minimized cost can be searched for by taking a path with a minimum A star score for each node.

[0092] In a global path planning according to one embodiment of the present invention, a rapidly exploring random tree (RRT) may be applied as a sampling technique for searching for an optimal path or the shortest path with a minimized path cost by connecting randomly sampled nodes from a start position of a robot to a target position. In the RRT, randomly sampled nodes are connected within an obstacle-free workspace, and the optimal path or the shortest path with a minimized path cost is searched for by connecting only parent nodes so that obstacles can be avoided while the path cost is minimized.

[0093] In global path planning according to various embodiments of the present invention, in addition to the A-star search as exemplified above, various search-based algorithms such as hybrid A-star search can be applied, and in addition to the RRT (rapidly exploring random tree), various sampling-based algorithms such as the RRT star or the PRM (probabilistic roadmap method) can be applied, and in addition to these search-based algorithms and sampling-based algorithms, an optimization-based algorithm that approaches global path planning as an optimization problem, such as the MPC (model predicted control), can be applied.

[0094] According to one embodiment of the present invention, a global path planning can generate a global path between a start position of a robot and a target position of the robot from a priori knowledge, such as spatial information (map) about a surrounding environment surrounding the robot, and in accordance with one embodiment of the present invention, a global path can be generated from a start position of the robot to a target position of the robot based on spatial information about the surrounding environment previously generated as a priori knowledge from simultaneous localization and mapping (SLAM) for simultaneously implementing localization for inferring a pose of the robot from spatial information about the surrounding environment surrounding the robot and mapping for generating spatial information about the surrounding environment surrounding the robot. In global path planning according to various embodiments of the present invention, a global path may be generated by considering collision avoidance with surrounding objects or obstacles that can be recognized from spatial information about a previously generated surrounding environment, or a global path may be generated without considering collision avoidance with surrounding objects or obstacles that can be recognized from spatial information about a previously generated surrounding environment. In this case, generation of a path for collision avoidance with such obstacles may be implemented from local path planning rather than global path planning.From this perspective, the local path generated from local path planning in this specification may comprehensively mean not only a local path for avoiding static and dynamic obstacles that are not recognized from spatial information about the surrounding environment, but also a local path for avoiding collisions with surrounding objects or obstacles that can be recognized from spatial information about the surrounding environment.

[0095] In one embodiment of the present invention, unlike global path planning that generates a global path from prior knowledge, local path planning may generate a local path for avoiding an obstacle detected based on observations from sensors of a robot observed in real time, for example, it may be generated in real time according to real-time observations from sensors of a robot following a global path.

[0096] FIG. 3 illustrates a diagram for explaining local path planning, which includes a global path as a candidate group of local paths set based on a preset rule, a group of lane paths set on both sides of a global path, a driving section including a local path selected from a cost function for each of the set group of lane paths, and a lane transition section including a bridge path that is incorporated as part of the local path to connect a pose of a robot on the selected local path and a current pose of the robot.

[0097] FIG. 4 illustrates a diagram for explaining a driving section and a lane transition section for a local pass generated from local pass planning, a local pass for avoiding static obstacles and dynamic obstacles, and a first type penalty set as a Gaussian distribution that gradually decreases along the obstacle and its periphery to provide a bias in the tendency to move away from the obstacle.

[0098] Figure 5 illustrates a diagram illustrating a first type penalty, which is set to a Gaussian distribution with a peak section extending as wide as the obstacle and gradually decreasing as it moves away from the peak section, to provide a bias toward moving away from the obstacle.

[0099] Figures 6a and 6b illustrate diagrams for explaining a second type penalty that induces the creation of a local pass as close to the global pass as possible, according to a second type penalty that tends to lead to a global pass within a range that can sufficiently avoid obstacles on the global pass.

[0100] FIGS. 7a to 7c illustrate a method for setting a local path outside the global path within a range sufficient to avoid an obstacle from the global path according to a first type penalty that provides a bias toward a tendency to move away from an obstacle (see FIG. 7a), and a method for setting a local path that returns to the global path again according to a second type penalty that provides a bias toward a tendency to move toward the global path from the position of the robot that has left the obstacle (see FIG. 7b). Different drawings are shown to illustrate the setting of local paths for each section that unfolds according to the robot's movement.

[0101] FIG. 8 illustrates a diagram for explaining a fourth type penalty for setting a local pass from a lane pass of a preferred side position among one side position and the other side position according to a fourth type penalty that is set unevenly for a group of lane passes of one side position and a group of lane passes of the other side position based on a global pass among a group of candidates for a local pass set on both sides of a global pass.

[0102] Figure 9 illustrates a table and diagram for explaining selecting a local pass from a lane pass (e.g., right side avoidance) of one side or the other side based on the global pass (Type 4 penalty) within a range where an obstacle can be sufficiently avoided (Type 1 penalty) or as close as possible to the global pass (Type 2 penalty) according to a cost function that synthesizes Type 1 to Type 4 penalties.

[0103] In local path planning according to one embodiment of the present invention, it is necessary to quickly generate a local path for avoiding obstacles in real time based on real-time observation observed from a robot's sensor. To this end, in local path planning according to one embodiment of the present invention, it is possible to prevent an excessive increase in the path cost predicted in the initial global path planning due to unexpected and excessive avoidance maneuvers for obstacle avoidance or collision avoidance while requiring relatively low computational cost. In addition, in local path planning according to one embodiment of the present invention, the concept of a collaborative robot interacting with a human can be introduced to implement predictable obstacle avoidance maneuvers or predictable collision avoidance maneuvers, and for example, safety accidents that may result from unexpected obstacle avoidance maneuvers or collision avoidance maneuvers of the robot can be prevented in advance. In other words, in local path planning according to one embodiment of the present invention, controllable local avoidance, controllable obstacle avoidance, or controllable collision avoidance maneuvers can be implemented, wherein, the controllable avoidance maneuver of the robot can mean that the robot avoidance maneuver is implemented within a preset range, while preventing an excessive increase in path cost due to unpredictable excessive avoidance maneuvers, and also preventing safety accidents due to unpredictable excessive avoidance maneuvers.

[0104] In local path planning according to one embodiment of the present invention, a cost function can be calculated for each candidate of a pre-set local path, and a local path in which the value of the cost function is minimized can be extracted. For example, in local path planning according to one embodiment of the present invention, a candidate of a local path that can be taken from the current position of a robot according to a pre-set rule can be calculated based on a rule, and a local path in which the cost function is minimized can be generated from the candidate of the local path calculated based on the rule in this way.

[0105] In one embodiment of the present invention, the rule for generating a candidate for a local path may be a configuration for implementing a controllable evasive maneuver or a controllable local evasive maneuver that can prevent excessive evasive maneuvers from excessively increasing the path cost predicted in the original global path planning or safety accidents caused by excessive evasive maneuvers or excessive evasive maneuvers that are beyond expectations by blocking excessive evasive maneuvers. For example, in local path planning according to one embodiment of the present invention, setting a candidate for a local path on a rule basis may mean that by limiting the degree of freedom (DOF) regarding the generation of the local path, a local path can be generated within a controllable or predictable range, thereby suppressing an increase in the path cost that is excessively beyond the original prediction and the occurrence of safety accidents caused by excessive obstacle avoidance or collision avoidance maneuvers that are beyond the prediction.

[0106] For example, in one embodiment of the present invention, as a rule for producing a candidate for a local path, a group of lane paths extending parallel to a global path generated from global path planning may be included, and more specifically, a plurality of groups of lane paths set along a width dimension or lane dimension intersecting the length dimension of the global path may be produced as candidates for the local path. The group of lane paths as candidates for the local path may include a plurality of lane paths extending parallel to the global path along the width dimension or lane dimension of the global path, including the global path, and as candidates for the local path, a group of lane paths set on both sides along the width dimension or lane dimension of the global path parallel to the global path may be produced as candidates for a group of local paths that can be taken from the current position of the robot.

[0107] In one embodiment of the present invention, generating a candidate for the local path may mean setting a group of lane paths along a global path generated through global path planning with a lane spacing taught in advance on both sides of the global path, for example, a plurality of lane paths may be set on both sides from the global path with a lane spacing taught in advance, and in one embodiment of the present invention, a plurality of lane paths may be set from a near location to a far location from the global path with a lane spacing taught in advance on both sides of the global path, including the global path, and for example, generating a group of lane paths as a candidate for the local path may mean generating a plurality of lane paths with a lane spacing set differently depending on the position of the robot, for example, a plurality of lane paths with a lane spacing set to evenly divide the width dimension or the lane dimension of the global path from the current position of the robot along the global path. A plurality of lane paths may be set by setting a lane spacing along the width dimension or lane dimension of the global path or by setting a lane spacing at a constant value, and in various embodiments of the present invention, the lane spacing for producing candidates for local paths may be taught differentially or uniformly depending on the current position of the robot following the global path.

[0108] According to one embodiment of the present invention, a local path generated through local path planning may be generated from a starting position where autonomous driving of the robot is initiated so as to follow a global path generated through global path planning from a starting position where autonomous driving of the robot is initiated, for example, so as to follow a global path generated through global path planning from the starting position where autonomous driving of the robot is initiated. From this consideration, candidates for the local path may include a group of lane paths set on both sides of the global path, including the global path. For example, candidates for the local path may include a pair of 1st, 2nd, ..., nth (where n is an integer) lane paths set on both sides of the global path or the 0th lane path, with the global path as a 0th lane path. For example, in one embodiment of the present invention, a group of lane paths forming candidates for the local path may include a pair of primary lane paths symmetrically set on both sides of the 0th lane path at a short distance from the 0th lane path, centered on a 0th lane path corresponding to a global path, a pair of secondary lane paths symmetrically set on both sides of the outer 1st lane path on both sides of the 0th lane path, and may include a plurality of pairs of similarly set lane paths, and may include a pair of nth lane paths set on both sides at a long distance from the 0th lane path.However, unlike the embodiment in which lane paths are not formed in pairs on both sides of the global path in various embodiments of the present invention, for example, the same number of lane paths are formed on both sides of the global path, in various embodiments of the present invention, the lane paths may be set in different numbers on one side and the other side of the global path, and the width dimension or lane dimension formed on one side of the global path and the width dimension or lane dimension formed on the other side of the global path may be formed asymmetrically along the global path which may be set to follow the shortest distance from the current position or the start position to the target position, and in various embodiments of the present invention, the lane paths may be set in different numbers on one side and the other side of the global path.

[0109] In local path planning according to one embodiment of the present invention, candidates for local paths are calculated based on a rule set in advance, and a local path for a robot to follow is selected or generated from among the candidates for local paths calculated in this way according to a cost function, thereby preventing in advance the risk of a safety accident due to excessive increase in the path cost predicted in the original global path planning due to excessive obstacle avoidance or excessive collision avoidance maneuvers that excessively deviate from the global path generated through global path planning, or due to excessive obstacle avoidance or collision avoidance maneuvers that deviate from the prediction. For example, in local path planning according to one embodiment of the present invention, setting a candidate for a local path in a rule-based manner may mean that a local path can be generated within a controllable or predictable range by limiting the degree of freedom (DOF) regarding the generation of a local path, thereby suppressing an increase in a path cost that excessively deviates from an initial prediction and an occurrence of a safety accident due to an excessive obstacle avoidance or collision avoidance maneuver that deviates from the prediction.For reference, through this specification, generating a candidate for a local pass based on a rule may mean that a candidate for a local pass can be set according to a rule set in advance (for example, setting a plurality of lane paths on both sides of a global pass so as to set an equal lane spacing along a width dimension or lane dimension intersecting the length of the global pass), and may not mean setting the local pass itself according to a rule set in advance, and as described below, in one embodiment of the present invention, the selection of a local pass or the generation of a local pass can generate an optimized lane path as a local pass according to the calculation of a cost function for each of a plurality of lane paths set based on a rule.

[0110] In local path planning according to one embodiment of the present invention, when selecting a local path from a plurality of local path candidates set along the width dimension or lane dimension of a global path, a penalty that gradually increases from a lane path at a close location to a lane path at a far location from the global path can be applied in calculating a cost function for the candidates of the local path so as to provide a bias toward a 0th-order lane path corresponding to the global path. In other words, a reward that gradually increases from a lane path at a far location to a lane path at a close location from the global path can be applied so as to provide a bias toward a 0th-order lane path corresponding to the global path among the candidates of the local path.

[0111] In one embodiment of the present invention, a cost function for each of a group of local path candidates forming a group of local path candidates can be derived from the sum of different types of penalty functions or reward functions, and a local path can be selected from among a plurality of local path candidates to optimize the cost function, more specifically, to minimize the total penalty of the cost function, or to maximize the total reward of the reward function. For example, providing a bias toward a 0th-order lane path corresponding to a global path among the local path candidates can be implemented by minimizing the cost function reflected in the form of a penalty (or penalty function) or a reward (or reward function) or by maximizing the reward function.

[0112] In local path planning according to one embodiment of the present invention, when selecting a local path from a plurality of local path candidates set along the width dimension or lane dimension of a global path, a penalty that gradually increases from a lane path at a short distance from the current position of the robot along the width dimension or lane dimension of the global path to a lane path at a long distance from the current position of the robot can be applied in calculating a cost function for the candidates of the local path so as to provide a bias for the current position of the robot. In other words, a reward that gradually increases from a lane path at a long distance from the current position of the robot to a lane path at a short distance from the current position of the robot can be applied so as to provide a bias toward the current position of the robot along the width dimension or lane dimension of the global path among the candidates of the local path.

[0113] In a local path planning according to one embodiment of the present invention, while generating a local path for obstacle avoidance or collision avoidance maneuver on a global path generated through global path planning, a local path for the robot to ultimately follow is selected or generated based on a cost function for each of a plurality of local path candidates calculated in a rule-based manner, and in calculating the cost function, a penalty or reward is set to provide a bias toward the global path, so as to induce the generation of a local path that is as close to the global path as possible within a range where obstacles on the global path can be sufficiently avoided, and an increase in path cost due to excessive avoidance maneuvers that excessively bypass the global path can be suppressed, and safety accidents can be prevented through predictable avoidance maneuvers. In addition, in local path planning according to one embodiment of the present invention, in calculating a cost function for each of a plurality of local path candidates calculated rule-based, a penalty or reward is set to provide a bias toward the current position of the robot, thereby inducing the generation of a local path that is as close to the current position of the robot as possible within a range that can sufficiently avoid obstacles on the global path, thereby suppressing an increase in path cost due to excessive avoidance maneuvers that excessively bypass the current position of the robot, and preventing safety accidents through predictable avoidance maneuvers.

[0114] In this way, in local path planning according to one embodiment of the present invention, a bias toward a global path and a bias toward a current position of the robot are provided together, but a relatively high weight (e.g., a relatively high penalty or a relatively high reward) is given to the bias toward the global path rather than the bias toward the current position of the robot, so that when the current position of the robot deviates from the global path generated through global path planning, a bias can be provided to follow the global path from the current position of the robot, and for example, when the starting position (corresponding to the current position) where autonomous driving of the robot begins deviates from the global path, a bias toward the global path from the starting position (the current position) can be provided, so that a local path can be generated so that the robot can follow the global path from the current position, and the local path of the robot can be generated so that the robot can drive by following the global path that is optimized by taking the path cost into account. In this way, in local path planning according to one embodiment of the present invention, a local path can be generated for obstacle avoidance or collision avoidance on a global path, and even if there is no obstacle on the global path, a local path can be generated from the current position of the robot that has left the global path toward the global path generated through global path planning. For example, if a penalty or reward of 0.9 scale is given to a bias toward the global path, a bias toward the current position of the robot is 0.A penalty or reward of 5 scales may be given, and thus, in local path planning according to one embodiment of the present invention, a weight of approximately twice that of a bias toward a global path may be given to a bias toward the current position of the robot.

[0115] In local path planning according to one embodiment of the present invention, a local path to be ultimately followed by a robot is selected or generated based on a cost function for each of a plurality of local path candidates calculated in a rule-based manner, and in calculating the cost function, an obstacle captured on a global path, for example, a static / dynamic obstacle recognized by a sensor of the robot at the current location of the robot, or a static / dynamic obstacle that can be recognized from spatial information about the surrounding environment surrounding the robot as prior knowledge for generating a global path in global path planning, can be avoided by providing a bias toward moving away from the obstacle, and a penalty or reward can be set to provide a bias toward moving away from the obstacle, thereby inducing the generation of a local path within a range in which the obstacle on the global path can be sufficiently avoided.

[0116] For example, in one embodiment of the present invention, a penalty of a Gaussian distribution can be set according to the distance from an obstacle following the width dimension of the global path or the lane dimension, and the penalty of the Gaussian distribution can be set so that the penalty gradually decreases along the Gaussian distribution according to the distance away from the obstacle while forming a peak at the location of the obstacle, and in one embodiment of the present invention, a penalty can be set so that the penalty decreases along the Gaussian distribution from the location away from the obstacle while forming a peak as wide as the obstacle along the width dimension of the global path or the lane dimension, and the peak section is formed by extending the peak of the Gaussian distribution as wide as the obstacle, and a relatively high penalty corresponding to the peak is given as wide as the obstacle, and even if the width of the obstacle is exceeded, the penalty gradually decreases along the Gaussian distribution. By imposing a decreasing penalty, a local path can be created with a sufficient margin of safety to sufficiently avoid the width of the obstacle, and in one embodiment of the present invention, a penalty of a Gaussian distribution can be set along the periphery of an obstacle that exceeds the width of the obstacle, taking into account the margin of safety, such as the transport load corresponding to the transport target of the robot protruding from the main body of the robot due to slipping, etc., or the steering radius according to the steering of the robot that bypasses the global path and follows the local path.For example, in one embodiment of the present invention, setting a penalty of a Gaussian distribution along the periphery of the obstacle may mean setting a penalty of a distribution that gradually decreases while following a Gaussian distribution while forming a peak section that extends flatly as far as the width of the obstacle, unlike a Gaussian distribution that has a peak at one location, as far as the width of the obstacle goes beyond the peak section corresponding to the width of the obstacle.

[0117] In a local path planning according to one embodiment of the present invention, a local path to be ultimately followed by a robot is selected or generated based on a cost function for each of a plurality of local path candidates calculated in a rule-based manner, and in calculating the cost function, a penalty or reward can be set in the calculation of the cost function to provide 1) a bias toward a tendency to move away from an obstacle recognized from the robot's current position, 2) a bias toward a global path, and 3) a bias toward the robot's current position, and the highest level of weight can be given to 1) the bias toward a tendency to move away from an obstacle, then 2) a middle level of weight can be given to the bias toward a global path, and finally 3) a lowest level of weight can be given to the bias toward the robot's current position. For example, in the calculation of a cost function for a group of lane paths as a candidate for a rule-based local path, if 1) a bias toward a tendency to move away from an obstacle is given a penalty of scale 4.5, 2) a bias toward a global path, and 3) a bias toward the current position of the robot may be given penalties of scale 0.9 and scale 0.5, respectively, and thus, in local path planning according to one embodiment of the present invention, 1) a bias toward a tendency to move away from an obstacle may be given a weight that is approximately 4.5 times as high as 2) a bias toward a global path, and 2) a weight that is approximately 9 times as high as a bias toward a global path.

[0118] For example, in the first penalty for providing a bias of a tendency to move away from the obstacle, the obstacle is an obstacle recognized from the current position of the robot at the current time point, that is, the current time point when local path planning is implemented, and may include, for example, an obstacle recognized from an observation observed from a sensor of the robot, and for example, an obstacle to which the first penalty is added may include an obstacle that can be recognized from spatial information about the surrounding environment as prior knowledge, and may include, for example, an obstacle that can be recognized from spatial information about the surrounding environment as prior knowledge for global path planning, and as described above, even if it is an obstacle that can be recognized from spatial information about the surrounding environment, spatial information based on the time point of generating spatial information may be, for example, spatial information at the time of mapping for generating spatial information about the surrounding environment, or mapping and Together, as spatial information at the time of SLAM that simultaneously implements localization for inferring the location of the robot, for example, it may be different from the spatial information at the present time, and accordingly, in various embodiments of the present invention, in global path planning, even if it is an obstacle that can be recognized from spatial information about the surrounding environment as prior knowledge, rather than generating a global path that takes into account avoidance of such obstacle, it is possible to generate a local path that takes into account avoidance of the obstacle in local path planning, and accordingly,The obstacle regarding the first penalty above is an obstacle captured on the global path, and may include, for example, an obstacle that could be considered in global path planning, but was not considered.

[0119] According to one embodiment of the present invention, a cost function for a group of lane paths as candidates for a local pass can be derived from the sum of different types of penalties or penalty functions as follows.

[0120] 1) A first-type penalty or first-type penalty function that adds a Gaussian-distributed penalty to the periphery of the obstacle along the width dimension or lane dimension of the global path for an obstacle recognized from the current position of the robot (or an obstacle captured in the global path or around the global path).

[0121] For example, for an obstacle on a 0th lane path corresponding to a global pass, a first-class penalty may be set for a pair of 1st lane paths set on both sides of the 0th lane path together with the 0th lane path, and further, a first-class penalty reduced according to a Gaussian distribution may be set for a pair of 2nd lane paths set on the periphery of the pair of 1st lane paths.

[0122] 2) A second type penalty that imposes a pre-set penalty according to the transition of a unit lane path along the width dimension of the global path or the lane dimension from the global path.

[0123] For example, a second penalty may be imposed by adding a preset second type scale to each unit transition corresponding to a unit lane path along both sides from a 0th lane path corresponding to a global pass, and for example, in one embodiment of the present invention, a second penalty may be imposed by adding a second type unit scale (e.g., 0.9) to each unit transition corresponding to a unit lane path. For example, a 0th lane path corresponding to a global pass may be assigned a 2nd type penalty of zero (0), a 2nd type unit scale may be assigned as a 2nd type penalty for a pair of 1st lane paths that are separated by a unit lane path or a unit transition from the 0th lane path corresponding to the global pass, and a 2nd type unit scale may be assigned as a 2nd type penalty for a 2nd lane path that is separated by a unit lane path or twice a unit transition from the 0th lane path corresponding to the global pass.

[0124] 3) A third type penalty that imposes a pre-set penalty according to the transition of a unit lane path along the width dimension of the global path or the lane dimension from the current position.

[0125] For example, in one embodiment of the present invention, a third type penalty may be imposed by adding a third type unit scale (e.g., 0.4) to each unit transition corresponding to a unit lane path. For example, a third-type penalty of zero (0) may be assigned to the current position of the robot or the 0th-order lane path closest to the current position, a third-type unit scale may be assigned to a pair of first-order lane paths that are one lane path or one transition away from the current position of the robot or the 0th-order lane path closest to the current position, and a third-type penalty of twice the third-type unit scale may be assigned to a second-order lane path that is one lane path or two transitions away from the current position of the robot or the 0th-order lane path closest to the current position.For example, as a candidate for a local pass, a 0th lane path corresponding to a global pass, a 1st lane path on both sides of the 0th lane path, and a 2nd lane path on both sides of the 1st lane path are set, and if the current position of the robot is closest to the outermost 2nd lane path, then zero (0) can be applied as a 3rd type penalty to the 2nd lane path closest to the current position of the robot, and a 3rd type unit scale can be applied as a 3rd type penalty to the 1st lane path that is away from the 2nd lane path closest to the current position of the robot by a unit lane path or a unit transition, and a unit lane path from the 2nd lane path closest to the current position of the robot For a 0th lane path that is twice the unit transition distance, a third-order penalty of twice the third-order unit scale may be imposed.

[0126] In one embodiment of the present invention, for a group of lane paths as candidates for a local path, a pair of first lane paths may be set symmetrically along both sides from a 0th lane path corresponding to a global path along the width dimension or lane dimension of the global path, and a pair of second lane paths may be set symmetrically at an outer location of the first lane path, and in this way, a pair of first lane paths, a pair of second lane paths, ..., nth lane paths may be set symmetrically and sequentially at an outer location of the 0th lane path, and more specifically, the pair of first lane paths may include a 1-1st lane path on one side and a 1-2nd lane path on the other side along the left and right sides of the 0th lane path. And, similarly, the pair of the secondary lane paths may include one 2-1 lane path and the other 2-2 lane path along both left and right sides of the 0-th lane path, and the pair of the n-th lane paths may include one n-1 lane path and the other n-2 lane path along both left and right sides of the 0-th lane path.

[0127] Regarding the calculation of the third type penalty (penalty or third type penalty function) from the first type penalty (penalty or first type penalty function), in the first type penalty (penalty or first type penalty function) that provides a bias toward a tendency to move away from an obstacle, a first type penalty that is differentially applied to a lane path set on one side of the 0th lane path and a lane path set on the other side may be applied depending on the location of the obstacle, and in the third type penalty (penalty or third type penalty function) that provides a bias toward the current location of the robot, a third type penalty that is differentially applied to a lane path set on one side of the 0th lane path and a lane path set on the other side may be applied depending on the current location of the robot. For example, in the first type penalty and the third type penalty, the first type penalty or the third type penalty may be differentially imposed on one lane path and the other lane path based on the zero-order lane path depending on the location of the obstacle or the current location of the robot, except in cases where the location of the obstacle is recognized on the zero-order lane path or the lane path closest to the current location of the robot is recognized as the zero-order lane path.

[0128] In one embodiment of the present invention, a cost function for each of a group of lane paths as candidates for the local pass can be calculated from a sum of different types of penalties, for example, can be calculated from a sum of different types of penalties, a first type penalty, a second type penalty, and a third type penalty. At this time, the weights for each of the penalties to be summed for calculating the cost function according to one embodiment of the present invention can be differentially assigned, and a relatively high weight can be assigned in the order of the first type penalty > the second type penalty > the third type penalty. More specifically, the weight of the first type penalty may be set to be approximately 4.5 times higher than the weight of the second type penalty, the weight of the first type penalty may be set to be approximately 9 times higher than the weight of the third type penalty, and the weight of the second type penalty may be set to be approximately 2 times higher than the weight of the third type penalty. For example, in one embodiment of the present invention, the first to third types of penalties may be imposed at a 4.5 scale, a 0.9 scale, and a 0.5 scale, respectively, where the 4.5 scale of the first type of penalty may be a penalty corresponding to a peak section of a Gaussian distribution corresponding to the width of an obstacle, and the 0.9 scale and the 0.5 scale of the second and third types of penalties may correspond to penalties of the second and third unit scales imposed per unit transition corresponding to a unit lane path.

[0129] In one embodiment of the present invention, the cost function may be calculated for each of a group of lane paths symmetrically set on both sides of the global path, including the global path, and may further include a fourth type penalty for suppressing the creation of a local minimum in the selection of a lane path so that the cost function is minimized while being asymmetrically applied to a lane path at one side and a lane path at the other side with respect to the global path, and in one embodiment of the present invention, the fourth type penalty is not equally applied to a lane path at one side with respect to the global path, and the fourth type penalty is equally applied to a lane path at the other side, and the weight of the fourth type penalty is greater than each of the first type penalty, the second type penalty, and the third type penalty. It can be low, for example, the fourth type penalty can be given the lowest weight among all the penalties considered in the calculation of the cost function, for example, the first type penalty to the fourth type penalty, and for example, unlike the first type penalty to the third type penalty, the fourth type penalty can prevent a convergence problem from occurring when a local minimum is formed on both sides of the global path in the minimization of the cost function, and for example, by imposing a differential penalty on the lane path of one side of the global path and the other side, a global maximum can be induced at one side of the global path, and for example, according to the setting of the fourth type penalty,In order to increase the predictability of the maneuver of a robot following a local path, unlike the first to third type penalties, a fourth type penalty can be set on a sufficiently small scale that can induce an imbalance in the lane path of one side and the other side of the global path. For example, in local path planning according to one embodiment of the present invention, a cost function can be derived from the sum of different types of first to fourth type penalties, and a lane path that can minimize the cost function according to the sum of the first to fourth type penalties can be selected as the local path for the robot to follow.

[0130] In a local path planning according to one embodiment of the present invention, a local path can be generated to avoid an obstacle (or an obstacle recognized from the current position of the robot) recognized along a global path generated from global path planning or the periphery of the global path, or a local path can be generated to follow the global path from the current position, and a group of lane paths parallel to the global path, including the global path, can be set as candidates for the local path for generating the local path, and a lane transition section can be set for transitioning the pose of the robot from the current position of the robot or from the lane path that the robot is currently following toward a lane path selected as the local path among the group of lane paths. For example, in local path planning according to one embodiment of the present invention, a local path may be generated for obstacle avoidance or global path following, and a lane transition section may be generated to change the pose of the robot from the current position of the robot or from the lane path that the robot is currently following to the local path (one lane path selected from a group of lane paths) to be followed by the robot to follow the local path. For example, the lane transition section of the robot may be set to a sufficient length to allow pose transition of the robot in consideration of the motion constraint of the robot.

[0131] In one embodiment of the present invention, the length of a lane transition section or a bridge path forming a lane transition section can provide sufficient cushioning (e.g., by forming a continuous smooth bridge path that excludes discontinuities that may cause abrupt pose changes) between the motion constraints of the robot or between different poses of the robot, and can be generated to have a length that is gradually extended according to the driving speed of the robot, for example. In other words, in local path planning according to one embodiment of the present invention, the length of a bridge path connecting a predetermined current pose of the robot and a pose of the robot on a pre-selected or pre-planned lane path can be extended by that amount according to the driving speed of the robot, so as to provide a more extended cushioning according to the driving speed of the robot. In other words, the length of the bridge path can be generated to have a length that is lengthened or shortened in conjunction with the driving speed of the robot.

[0132] In this specification, for the convenience of understanding, one lane path selectively selected from a group of lane paths extending in parallel along a global path is referred to as a local path. However, more precisely, the local path according to one embodiment of the present invention may comprehensively mean all of a series of paths generated through local path planning. For example, in one embodiment of the present invention, the local path generated through local path planning may include a driving section including one lane path selectively selected from a group of lane paths extending in parallel along a global path, and a lane transition section for transitioning the pose of the robot to follow the selected lane path, or for changing the pose of the robot from the current position or the lane path followed by the current position toward the selected lane path. As described above, among a group of lane paths extending in parallel along a global path, Even in instances where a change in the pose of the robot is not required, for example, a lane transition section for connecting the candidate set of local paths and the current position of the robot can be maintained according to the cost function calculation for each of a set of candidates of local paths including a global path and a lane path extending parallel to the global path, even if the pose of the robot following the current global path is maintained, and for example, a local path generated from local path planning so that local path planning can be implemented from the same algorithmic process in any instance,It may include a lane path selected from a cost function among a global path and a group of lane paths extending parallel to the global path, or a driving section including the selected lane path, and a bridge path or a lane transition section including the bridge path for connecting the lane path selected from the cost function and the current position of the robot.

[0133] In one embodiment of the present invention, the lane transition section is for forming a bridge path from the current position of the robot or the lane path currently being taken by the robot to the lane path selected from the cost function among a group of lane paths (1st lane path to nth lane path) extending in parallel along the global path (0th lane path), and for example, when the current position of the robot or the lane path currently being taken by the robot and the lane path selected from the cost function do not match each other and the pose of the robot needs to be changed between different lane paths, the bridge path may include a bridge path that connects the poses of these robots, and in various embodiments of the present invention, even when the current position of the robot or the lane path currently being taken by the robot and the lane path selected from the cost function match each other, the bridge path that connects them A path) can be generated, for example, in local path planning according to various embodiments of the present invention, a global path and a group of lane paths extending parallel to the global path are set, a lane path that minimizes a cost function is selected from among the set group of lane paths, and a bridge path that connects the current location of the robot or the lane path that the robot is currently taking and the selected lane path can be generated.That is, the bridge path can provide a path connecting the current position of the robot or the lane path currently being taken by the robot and the lane path selected from the cost function, and motion planning for the bridge path can be implemented in the local path planning so that the pose of the robot including the heading or rotation of the robot at each position along the bridge path can be planned.

[0134] In this way, in local path planning according to one embodiment of the present invention, a bridge path that connects a lane path selected from a cost function and a current robot position can be generated, and in the local path planning, the pose of the robot can also be planned at the position of each robot along the bridge path, and for example, a bridge path that connects the current position of the robot and the position of the robot designated at both end positions of the bridge path corresponding to the lane path selected from the cost function and the heading or rotation of the robot can be generated. For example, in an instance where the current robot is following a global path, the local path planning may set a group of candidates for local paths that extend parallel to the global path, and a lane path selected by considering a cost function from a group of lane paths that extend parallel to the global path and the global path from the global path that the current robot is following may be parallel to the global path that the current robot is following, and accordingly, the pose of the robot may be formed as a heading or rotation that is parallel to each other between the current position of the robot or the global path that the robot is currently following and the lane path selected by considering the cost function on a bridge path that connects the lane path selected by considering the cost function and the lane path that the current position of the robot or the global path that the robot is currently following may be parallel to each other, and in order to avoid or collide with an obstacle recognized at the current position of the robot while connecting the poses of the heading or rotation of the robot that are parallel to each other,For example, a second lane path, rather than a 0th lane path corresponding to a global path, may be selected from a cost function, and in this case, in local path planning, a bridge path may be generated so that the poses of robots at opposite end positions that are parallel to each other can be smoothly connected while the robot can take a second lane path that bypasses the global path in order to avoid an obstacle recognized on the global path or the global path that the robot is currently taking, while connecting the second lane path that bypasses the global path from the robot's current position or the global path that the robot is currently taking, so that the poses of the robots with the same heading or rotation can be formed at both end positions.

[0135] In local path planning according to one embodiment of the present invention, the poses of the robot at both end positions between the current position of the robot and the lane path selected from the cost function may be predetermined (for example, the current position of the robot corresponds to prior knowledge of the local path planning at the current position) or may be pre-planned, and for a bridge path that connects the predetermined or pre-planned positions of the robot and the poses of the robot at both end positions including the heading and rotation of the robot, the local path planning may generate a bridge lane that smoothly connects the predetermined or pre-planned poses of the robot at these both end positions.

[0136] For example, in local path planning according to one embodiment of the present invention, generating a bridge path that smoothly connects the poses of the robot at the predetermined or planned end positions may mean, for example, not generating a discontinuity or a discontinuous singularity along the bridge path, and in local path planning according to one embodiment of the present invention, generating a smooth bridge path that does not have a discontinuous curvature such as a sharp corner along the bridge path, or in other words, generating a bridge path that has a smooth curve with continuous curvature.

[0137] For example, in local path planning according to one embodiment of the present invention, the bridge path may be generated in the form of a polynomial of degree 3 or higher, or the bridge path may be generated from a heuristic estimation to avoid discontinuous curvatures such as sharp corners, and a smooth bridge path may be generated through smoothing, such as removing discontinuities on the generated bridge path, for example, removing discontinuous curvatures.

[0138] In local path planning according to one embodiment of the present invention, a smooth bridge path that does not have a discontinuous curvature or includes a continuous curvature can be generated by taking into consideration the problem of the motion constraint of a robot that has difficulty in following the discontinuous curvature of a bridge path and the resulting decrease in the driving stability of the robot, and the problem of the deterioration of the posture stability of a transport object carried by the robot due to the robot's rapid steering to follow the discontinuous curvature of the bridge path.

[0139] In one embodiment of the present invention, unlike global path planning for generating a global path from a starting position or a starting pose to a goal position or a goal pose based on prior knowledge such as spatial information about the surrounding environment, for example, in local path planning implemented in real time based on the progress of a time step along a robot's driving path or based on static / dynamic obstacles captured in observations from sensors of the robot at the current position, it is necessary to quickly implement a local path for obstacle avoidance or collision avoidance in real time so that the robot's driving is not delayed, and for example, in local path planning for reducing computational cost and preventing work delays of the robot due to computational cost or computational time and shortening work time, candidates for local paths are set rule-based to include global paths and A group of parallel lane paths are set as candidates for local paths, and a cost function is calculated for each lane path to optimize the cost function, for example, a lane path for which the cost function is minimized is selected as a local path for the robot to follow, while a relatively low computational cost is used to select a local path (a lane path selected from the cost function) and the current position of the robot, or a pose of the robot on a lane path selected from the cost function among a group of lane paths,By generating a bridge path that connects the current poses of the robot in a smooth form, a local path for avoiding static / dynamic obstacles can be generated with a relatively low computational cost, while the stability of the robot's driving according to the motion constraints or the posture stability of the transport object carried by the robot can be prevented from being damaged by the bridge path that smoothly connects the local path (lane path selected from the cost function) generated with such a low computational cost and the current position or current force of the robot. As explained above, by selecting a lane path for the robot to follow from a group of lane paths set in a controllable range, for example, rule-based, it is possible to prevent an excessive increase in the path cost estimated in the initial global path due to an excessive obstacle avoidance or collision avoidance maneuver that goes beyond prediction, or a safety accident from occurring due to an excessive obstacle avoidance or collision avoidance maneuver that goes beyond prediction. there is.,

[0140] In a local path planning according to one embodiment of the present invention, the cost function for each of a global path and a group of lane paths extending parallel to the global path as candidates for the local path can be calculated by adding 1) a first-type penalty function for providing a bias of a tendency to move away from an obstacle, 2) a second-type penalty function for providing a bias of a tendency to move toward the global path, and 3) a third-type penalty function for providing a bias of a tendency to move toward the current position of the robot, and in the calculation of such a cost function, as each of the first to third-type penalties for providing 1) bias or 3) bias, a first-type penalty having a peak section corresponding to the width of the obstacle as set in advance and decreasing in a Gaussian distribution while leaving the obstacle, and each of the global paths and the current position of the robot A cost function can be calculated from the sum of the second and third type penalties, which are given a unit scale (the second type unit scale and the third type unit scale) for each unit transition corresponding to a unit lane path from the lane path closest to the current position of the robot, and in calculating the cost function for selecting a local path according to one embodiment of the present invention, the cost function can be calculated from a simple operation with a relatively low computational cost, and for example, it is an operation mainly centered on summation, such as the sum of the first to third type penalties, and for example,A cost function for selecting a lane path for a robot to follow among a set of lane paths can be derived from the relatively reduced computational cost from operations that are mainly centered on summation rather than multiplication.

[0141] Through this specification, global path planning is established along the entire driving path of the robot from the start position or start pose of the robot to the goal position or target pose of the robot; And local path planning for generating a local path including a bridge path for changing the pose of the robot from the current position of the robot or from the lane path currently being followed by the robot toward the selected local path to be followed by the robot, wherein a group of lane paths are set as local path candidates on both sides of the global path based on a previously set rule, and a cost function is calculated for each lane path forming the set local path candidate group, for example, the cost function is calculated from the sum of a first type penalty, a second type penalty, a third type penalty, and a fourth type penalty for providing different types of bias, and a bridge path for changing the pose of the robot from the current position of the robot or from the lane path currently being followed by the robot toward the selected local path to be followed by the robot;As for the above-described global path planning and local path planning, the global path planning and local path planning described above can be implemented from a calculation unit for performing necessary calculations to perform the autonomous driving system of the robot implementing the controllable local avoidance of the present invention and controlling or controlling the entire autonomous driving system, and for the convenience of understanding in several places of this specification, even if the calculation unit is not specified as a subject for performing the autonomous driving system of the robot implementing the controllable local avoidance of the present invention or as a subject for calculation and / or control to implement the autonomous driving system of the robot described above, as a subject for implementing the technical configuration described above, the calculation unit of the present invention can perform actions and functions necessary to perform and implement the corresponding technical configuration.;

[0142] Although the present invention has been described with reference to the embodiments shown in the attached drawings, these are merely exemplary, and those skilled in the art to which the present invention pertains will understand that various modifications and equivalent other embodiments are possible therefrom.

[0143] The present invention can be applied to autonomous driving systems for robots and industrial fields related to autonomous driving for robots.

Claims

A computation unit for implementing global path planning for generating a global path from the robot's starting position to the robot's target position, and local path planning for generating a local path narrower than the global path centered on the robot's current position. The above operation unit is, In the above local path planning, a group of lane paths are set that include the global path and extend parallel to the global path along the width dimension or lane dimension of the global path, An autonomous driving system for a robot that calculates a cost function for each of a set of lane paths and generates a lane path, which is selectively selected from among the set of lane paths based on the calculated cost function, as a local path for the robot to follow. In the first paragraph, The above operation unit is, Rule-based setting of candidates for local passes by forming a group of lane passes as candidates for local passes along the pre-set lane spacing on both sides of the above global pass, An autonomous driving system for a robot, characterized in that a lane path to be followed by the robot is selected from candidates of the above-described regional paths. In the second paragraph, The above operation unit is, By setting the above group of lane paths symmetrically on both sides of the above global pass, An autonomous driving system for a robot, characterized in that the above-mentioned global pass is a 0th lane pass, a pair of 1st lane passes is set symmetrically on both outer sides of the 0th lane pass, and a pair of 2nd lane passes is set symmetrically on both outer sides of the pair of 1st lane passes. In the first paragraph, The above operation unit is, An autonomous navigation system for a robot, characterized in that it generates a local path to avoid obstacles recognized from the current location of the robot or to follow a global path from the current location of the robot. In the first paragraph, The above cost function is, 1) A first type penalty to provide a bias towards moving away from the perceived obstacle from the robot's current position; 2) a second type penalty to provide a bias towards global passes; and 3) An autonomous driving system for a robot, characterized in that it is derived from a penalty that provides at least one bias among the third type penalties for providing a bias in the tendency to move toward the current position of the robot. In paragraph 5, The above cost function is, An autonomous driving system for a robot, characterized in that it is calculated from the sum of the first type penalty, the second type penalty, and the third type penalty. In paragraph 6, The weights of the above first type penalty, second type penalty and third type penalty are as follows: An autonomous driving system for a robot, characterized in that penalties are imposed in the order of Type 1 penalty > Type 2 penalty > Type 3 penalty. In paragraph 6, According to the relationship between the weights of the first, second and third types of penalties, The above operation unit is, Generate a global path and an adjacent local path within a sufficient range for obstacle avoidance or collision avoidance, or An autonomous navigation system for a robot, characterized in that it generates a local path toward a global path from the current position of the robot within a sufficient range for obstacle avoidance or collision avoidance. In paragraph 6, The above operation unit is, For the second type penalty and the third type penalty of the second type penalty for providing a bias toward the global pass in the above 2) and the third type penalty for providing a bias toward the current position of the robot in the above 3), respectively, According to the relationship between the second type unit scale and the third type unit scale given to each unit transition corresponding to the unit lane pass, An autonomous navigation system for a robot, characterized in that it generates a local path toward the global path from the current position of the robot outside the global path. In paragraph 5, The above cost function is calculated for each of a group of lane passes set on both sides of the global pass, including the global pass, An autonomous driving system for a robot, characterized in that it further includes a fourth type penalty for suppressing the creation of a local minimum in the selection of a lane pass so that a cost function is minimized while imbalancedly applying a lane pass of one side and a lane pass of another side centered on the above global pass. In Article 10, A robot autonomous driving system characterized in that a type 4 penalty is not uniformly applied to lane passes at one side of the above global pass, and a type 4 penalty is uniformly applied to lane passes at the other side, but the weight of the type 4 penalty is lower than each of the type 1 penalty, the type 2 penalty, and the type 3 penalty. In the first paragraph, The above operation unit is, An autonomous driving system for a robot, characterized in that it generates a local path toward the global path from a starting position where autonomous driving of the robot is initiated or a position estimated with respect to the starting position. In the first paragraph, The above operation unit is, An autonomous driving system for a robot, characterized in that, for a group of lane passes as candidates for the above-mentioned local pass, a first type penalty is added that includes a peak section corresponding to the width of an obstacle and decreases in a Gaussian distribution as it goes beyond the width of the obstacle or the peak section. In the first paragraph, The above operation unit is, An autonomous driving system for a robot, characterized in that a second penalty of a second type unit scale is added to each unit transition corresponding to a unit lane path from the global pass for the group of lane paths as candidates for the local pass. In the first paragraph, The above operation unit is, An autonomous driving system for a robot, characterized in that, for the group of lane paths as candidates for the above-mentioned local pass, a third-class penalty of a third-class unit scale is added for each unit transition corresponding to a unit lane path from the current position of the robot or the pass lane closest to the current position of the robot. In the first paragraph, The above operation unit is, Generate a bridge path to connect the current location of the robot and a lane path selected from a cost function among the group of lane paths as candidates for the local path, An autonomous driving system for a robot, characterized in that the generated bridge pass is incorporated into the local pass together with the lane pass selected from the cost function. In Article 16, The above operation unit is, An autonomous driving system for a robot, characterized in that it generates a local path including a lane path selected from the cost function from the local path planning and the bridge path connected to the lane path selected from the cost function. In Article 16, The above operation unit is, An autonomous driving system for a robot, characterized in that it implements motion planning that plans the pose of the robot according to each position along the bridge path so as to connect the pose of the robot determined in advance with respect to the current position of the robot forming the positions at both ends of the bridge path and the pose of the robot planned in advance on the lane path selected from the cost function. In Article 18, An autonomous driving system for a robot, characterized in that the pose of the robot on the bridge path planned from the above-mentioned local path planning includes the position of the robot and the heading or rotation of the robot. In Article 18, The above operation unit is, Generate a bridge path connecting the current pose of the robot forming the two end positions of the bridge path and the pose of the robot on the lane path selected from the cost function, An autonomous driving system for a robot, characterized in that it generates a bridge path that smoothly connects the poses of the robot at the two end positions so that no discontinuity or singularity is formed along the bridge path. In Article 20, The above operation unit is, An autonomous driving system for a robot, characterized in that it generates a bridge path that does not include a discontinuous curvature along the bridge path. In Article 21, The above operation unit is, An autonomous navigation system for a robot, characterized in that it generates a bridge path including a curved segment having a continuous curvature along the bridge path. In Article 21, The above operation unit is, To avoid including discontinuous curvature, the bridge path is generated in the form of a polynomial of degree 3 or higher, or Generate the bridge path from a heuristic estimate to avoid discontinuous curvature, or An autonomous navigation system for a robot, characterized in that the bridge path is generated from smoothing for discontinuous curvature. In Article 18, The above operation unit is, Generate a bridge path connecting the current pose of the robot forming the two end positions of the bridge path and the pose of the robot on the lane path selected from the cost function, An autonomous driving system for a robot, characterized in that it generates a bridge path of an extended or shortened length in conjunction with the driving speed of the robot.

Images