Layered gait planning method for legged robot in complex environment

Abstract

The application discloses a layered gait planning method for a legged robot in a complex environment, which is divided into outer layer planning and inner layer planning. The outer layer planning determines the step direction of each leg, and the inner layer planning determines the foot trajectory according to the step direction. Meanwhile, the outer layer planning and the inner layer planning are both composed of experience planning and learning planning. The experience planning plans corresponding actions according to a system model, and the learning planning plans by using a reinforcement learning method. The experience planning and the learning planning are weighted and fused to obtain a final planning result. The fusion weight gradually deviates to the learning planning with the training of the reinforcement learning. The outer layer experience planning adopts a ZMP method, and increases the stability margin by changing the position of the supporting leg. The inner layer planning adopts a cycloid as the foot trajectory, and determines the parameters of the elliptical trajectory according to the ZMP stability margin and the motion direction.

Description

Technical Field

[0001] The present invention belongs to the field of robot motion planning, and in particular relates to a hierarchical gait planning method for a legged robot in a complex environment. Background Art

[0002] Currently, common legged robot gait planning uses walking or diagonal gaits (two legs on a diagonal line are lifted or lowered at the same time). The robot lifts and lowers its legs according to a fixed pattern without considering terrain conditions. Alternatively, the robot uses a diagonal gait as a basis and changes the stride size according to terrain conditions, without changing the order in which the legs are lifted, nor does it selectively lift a leg based on terrain conditions. However, real quadrupeds do not have a fixed gait in complex terrain environments. Whether each leg is lifted or lowered is temporarily determined based on the terrain conditions and the body's state. The gaits planned by these methods appear mechanical, with poor adaptability to movement in complex terrain, low efficiency, and low stability. Summary of the Invention

[0003] In response to the problems existing in current legged robot motion planning methods, the present invention proposes a hierarchical gait planning method for legged robots in complex environments, which enables the robot to pass through different terrain environments with random gaits, thereby improving the efficiency and flexibility of movement.

[0004] The technical solution to realize the present invention is: a hierarchical gait planning method for a legged robot in a complex environment, characterized by: a random gait facing complex terrain, without presetting the order of each leg's step, but independently determining the direction of each leg's next step based on the robot state and the environment state; the entire gait planning adopts a hierarchical planning method, the outer layer planning is used to determine the direction of the leg step, and the inner layer planning is used to generate the foot end motion trajectory; both the outer layer planning and the inner layer planning are composed of two parts: experience planning and learning planning, which are then fused; after initialization, the robot performs state observation and determines whether to perform outer layer planning based on the observation results; outer layer planning is performed only when a swinging leg enters the support phase. Specifically, the following steps are included:

[0005] Step 1: Use sensors to obtain the robot and environmental status, and determine whether the foot contact status has changed. If the foot contact status has changed or all four legs of the robot are in contact with the ground, jump to step 2; otherwise, jump to step 3.

[0006] Step 2: Based on the acquired environment status and robot status, perform outer layer planning, select the foot end that needs to change position and plan its leg direction, then jump to step 3.

[0007] Step 3: Based on the changes in the foot contact state and the outer layer planning results, perform inner layer planning to generate the foot movement trajectory.

[0008] Step 4: Based on the planned foot end motion trajectory, the inverse kinematics method is used to determine the joint motion trajectory, and the action is completed through the actuator.

[0009] Step 5: Update the fusion coefficient of the experience planning and learning planning of the inner and outer planning.

[0010] Compared with existing technologies, this invention offers significant advantages: the legged robot adopts a random gait, which improves its flexibility and efficiency, and enhances its adaptability to complex terrain. Both inner and outer planning utilize a fusion of empirical and learning planning, reducing the complexity of motion planning. A progressive update strategy accelerates training and mitigates the adverse effects of uncertainty at the beginning of training. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is an overall flow chart of the present invention.

[0012] Figure 2 It is a schematic diagram of outer layer motion planning of the present invention.

[0013] Figure 3 It is a schematic diagram of the stepping direction of the present invention. DETAILED DESCRIPTION

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0016] The technical solutions between the various embodiments of the present invention can be combined with each other, but they must be based on the fact that ordinary technicians in this field can implement them. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0017] The following will further introduce the specific implementation methods, as well as the technical difficulties and inventive points of this invention in combination with this design example.

[0018] The quadruped robot walks in a random gait, where the left front leg steps in the direction of Figure 2 As shown by Figure 2 As can be seen, each leg has nine stepping directions. When a quadruped robot walks, if one or more legs touch the ground from the swing phase, the robot's foot contact state changes. At this point, external motion planning is required based on the overall motion instructions and state information. This determines which leg needs to transition from the support phase to the swing phase and selects a stepping direction from among the nine motion directions (including stationary movement).

[0019] Empirical planning uses the zero-moment point to determine the leg that needs to change its foot position and the direction of its stride. First, the distance from the zero-moment point to each side of the support polygon is calculated. The two support points with the largest distance are then found. The positions of these two points in the direction of the center of mass velocity are then compared. The support point with the smaller distance is the foot that needs to change its position. The direction of the stride is determined based on the direction of movement of the zero-moment point.

[0020] After planning the step direction, the specific location of the next footstep and the maximum leg lift height need to be determined. This is used to generate a complete foot trajectory. Based on this foot trajectory, inverse kinematics is then used to calculate the motion trajectory of each joint. Finally, the joint motion commands are sent to the joint actuators, completing the entire motion cycle.

[0021] The present invention provides a method for layered gait planning of a legged robot in a complex environment, and the specific steps are as follows:

[0022] Step 1: Update the status and obtain the current center of mass motion instruction to determine whether the foot contact status of each leg has changed:

[0023] 1-1) Obtain the current state of the robot and the environment, and calculate the zero moment point (ZMP).

[0024] 1-2) Compare the current foot contact state with the previous state to determine whether there has been a change, and select the next step based on whether the state has changed.

[0025] Step 2: If the foot contact state changes, perform outer trajectory planning based on the state and select the next step direction for the leg in the stance phase:

[0026] 2-1) Based on the movement command and ZMP, determine the next step direction of each leg.

[0027] 2-2) Based on the current state, the movement direction of each leg is determined using a discrete reinforcement learning algorithm.

[0028] 2-3) Combine the leg movement directions determined in the above two steps to determine the final movement direction of each leg.

[0029] Step 3: Based on the changes in the foot contact state and the results of the outer motion planning, perform inner trajectory planning to plan the foot's motion trajectory:

[0030] 3-1) Determine the foot trajectory based on the outer-layer planned direction of each foot and the current foot position. If the foot must transition from ground contact to airborne, plan the foot trajectory parameters. If the foot remains airborne, continue executing the current trajectory. If the foot remains in the support phase, plan the foot trajectory based on the robot's center of mass trajectory.

[0031] 3-2) Determine the foot end position at the next time point through a continuous reinforcement learning algorithm.

[0032] 3-3) A progressive fusion strategy is used to fuse the foot end positions determined in the first two steps to determine the final foot end position a. The fusion relationship is:

[0033] a=μa p +(1-μ)a m

[0034] Among them, a p is the foot end position given by the reference trajectory, a m is the foot end position given by the intelligent model, and μ is the fusion coefficient.

[0035] Step 4: Based on the planned foot end motion trajectory, the inverse kinematics method is used to determine the joint motion trajectory, and the action is completed through the actuator:

[0036] 4-1) Based on the foot trajectory, the inverse kinematics method is used to calculate the motion trajectory of each joint.

[0037] 4-2) Send the joint trajectory to the actuator for execution.

[0038] Step 5: Update the learning planning model according to the current status of the robot and environment, and adjust the fusion parameters.

Claims

1. A hierarchical gait planning method for a legged robot in a complex environment, characterized by: The random gait for complex terrain does not predetermine the order of each leg's steps. Instead, the robot independently determines the direction of each leg's next step based on the robot's state and the state of the environment. The entire gait planning adopts a hierarchical planning approach, with the outer layer used to determine the direction of the step and the inner layer used to generate the foot's motion trajectory. Both the outer and inner layers consist of two parts: experience planning and learning planning, which are then integrated. After initialization, the robot performs state observation and determines whether to proceed to the outer layer planning based on the observation results. Outer planning is performed only when the swing leg enters the stance phase, which includes the following steps: Step 1: Use sensors to obtain the robot and environment status and determine whether the foot contact state has changed. If the foot contact state has changed or all four legs of the robot are in contact with the ground, jump to step 2; otherwise, jump to step 3. Step 2: Based on the acquired environment and robot states, perform outer layer planning, select the foot that needs to change position, and plan its step direction, as follows; 2-1) Based on the motion command and the current state, determine the leg that needs to change its foot position and its step direction so that the center of the support polygon moves in the direction of the robot's motion; 2-2) Based on the current state, determine the direction of each leg's movement using a discrete reinforcement learning algorithm; 2-3) The step directions determined in the first two steps are combined to determine the final step direction for each leg; Then skip to step three; Step 3: Based on the changes in the foot contact state and the outer planning results, perform inner planning to generate the foot motion trajectory. The specific steps are as follows: 3-1) Determine the foot trajectory based on the outer layer's planned direction of each leg and the foot's current position. If the foot is currently in contact and does not need to be lifted, then plan the foot trajectory parameters. If the foot is in a swinging state, continue executing the current trajectory. 3-2) Determine the position of the foot at the next time point through a continuous reinforcement learning algorithm; 3-3) Using a progressive fusion strategy to fuse the foot end positions determined in the first two steps to determine the final foot end position; Step 4: Based on the planned foot end motion trajectory, the inverse kinematics method is used to determine the joint motion trajectory, and the action is completed through the actuator; Step 5: Update the fusion coefficient of the experience planning and learning planning of the inner and outer planning.

2. The hierarchical gait planning method for a legged robot in a complex environment according to claim 1, characterized in that: In step 1, the robot and environment status are acquired through sensors to determine whether the foot contact status has changed. The specific steps are as follows: 1-1) Update the current robot state and environment state and calculate the zero moment point; 1-2) Compare the foot contact state at the current moment with the previous moment to determine whether there has been a change.

3. The hierarchical gait planning method for a legged robot in a complex environment according to claim 1, characterized in that: The fusion relationship of step three is: a=μa p +(1-μ)a m Among them, a p is the foot end position given by empirical planning, a m is the foot position given by the learning plan, μ is the fusion factor, its initial value is 1, and it gradually decreases to 0 as the training progresses. Its changing rule is: μ = e -t , t represents the training process.

4. The hierarchical gait planning method for a legged robot in a complex environment according to claim 1, characterized in that: There are nine step directions for each leg, namely front, left front, left, left back, back, right back, right, right front, and stay in place.

5. The hierarchical gait planning method for a legged robot in a complex environment according to claim 1, characterized in that: The inner layer planning uses the elliptical trajectory as the reference trajectory, and the planning only changes the parameters of the elliptical trajectory.

6. The hierarchical gait planning method for a legged robot in a complex environment according to claim 1, characterized in that: The environmental state includes the undulation and softness of the terrain, and the robot state includes the velocity of the robot's center of mass, joint positions and velocities, and the contact force between the foot and the ground.

Images