737 results about "Legged robot" patented technology

The invention discloses a quadruped robot motion control method based on adaptive deep reinforcement learning. The method comprises the following steps: S1, determining a network model, a composite reward function, a state space and a bionic action generation mechanism of a simulation training environment; the state space provides environment information input, the network model processes the input information and generates a decision, the bionic action mechanism executes a specific decision behavior, and the composite reward function evaluates a behavior effect and optimizes a decision direction; s2, constructing a simulation training environment of the quadruped robot, wherein the simulation environment comprises quadruped robot model information and simulation environment information; s3, training the network model by using a deep reinforcement learning algorithm based on robot model information and simulated environment information to obtain a trained motion control strategy; and S4, verifying the feasibility of utilizing the trained motion control strategy by controlling the motion of the quadruped robot in a real environment. According to the invention, the self-adaptive capability to a complex environment is obviously improved.

A robot in accordance with at least some embodiments of the present technology includes a neck region through which the robot is configured to transmit and receive light. The robot defines a robot height and includes a body, a head carried by the body, and a neck extending between the head and the body along the robot height. The robot further includes a structural neck frame at a periphery of the neck in a plane perpendicular to the robot height. The structural neck frame defines windows distributed around the periphery of the neck. The robot further includes a sensor disposed at least partially within the structural neck frame. The sensor includes a laser configured to transmit light via the windows and a detector configured to receive light via the windows. A field of view of the sensor extends at least 320 degrees around the neck.

A balance control method for a wheel-legged robot is provided. The robot includes a moving wheel, n links, and n rotating joints, the moving wheel being connected to a first link through a first rotating joint of the n rotating joints, and the n links being connected in series through n−1 rotating joints other than the first rotating joint, n being a positive integer greater than 1. The method includes: acquiring a state quantity of the wheel-legged robot at a first moment; determining dynamics model parameters according to a dynamics equation of the wheel-legged robot and the state quantity at the first moment; establishing a sliding surface according to the state quantity at the first moment; calculating rotation torques of the n rotating joints according to the sliding surface and the dynamics model parameters; and controlling the rotating joints according to the rotation torques of the n rotating joints.

The invention discloses a self-adaptive gait control method, system and device for a quadruped robot in a limited space and a medium, and belongs to the technical field of gait control. Identifying terrain categories through a classifier and acquiring initial gait parameters from a gait library; adjusting the gait parameters according to the terrain gradient and the roughness to obtain optimized gait parameters; calculating stability indexes and fusing scores; when the score is lower than a threshold value, an adjustment strategy is determined according to an instability mode to correct gait parameters; a foot end track is generated and control is executed; and the executed information is fed back to terrain perception to form a closed loop. According to the method, the terrain recognition accuracy is improved through multi-modal information fusion, the adaptive mapping from the terrain to the gait is established to achieve automatic gait switching and parameter optimization, the stability is evaluated through multi-index fusion, a targeted adjustment strategy is selected according to an instability mode, the adaptive gait control problem under the limited space and the complex terrain is solved, and the adaptive gait recognition accuracy is improved. And the movement efficiency and the stability are improved.

The invention belongs to the related technical field of mobile robots, and discloses a rolling-walking composite motion control method and device for a double-wheel-foot robot, and the method comprises the steps: (1) simplifying the rolling motion of wheels of the double-wheel-foot robot into the sliding motion of a sliding block; (2) constructing a system kinetic equation of the double-wheel-foot robot based on the simplified kinetic model; constructing a constraint equation based on the contact state of the two legs and the ground; discretizing the system kinetic equation, the cost function and the constraint equation, and converting model prediction control problems corresponding to the system kinetic equation, the cost function and the constraint equation into nonlinear problems for numerical solution so as to complete leg motion planning; (3) calculating the rotation speed of the wheel based on the leg motion planning result; and (4) based on the rotation speed of the wheels and the leg motion planning result, a whole-body control algorithm is adopted to calculate and obtain moment instructions of leg joints and the wheels. According to the invention, the state space dimension is reduced, and the wheel-ground contact constraint is simplified.

The invention belongs to the technical field of robot control, and provides a quadruped robot fall-down self-recovery control method and system under complex terrains. The method comprises the following steps: extracting a tumble self-recovery strategy feature from privilege information subjected to normalization processing by using a teacher network, so as to provide an optimization reference for a student network; learning strategy characteristics simulating a teacher network by using a student network so as to process the normalized historical observation information of the ontology state information; after the strategy converges, the Actor network is used for processing the output characteristics of the student network and the real-time observation information of the normalized body state information, the next state observation information, the reward signal, the cost feedback and the task end mark are obtained through interaction with the simulation environment, and the action of the quadruped robot is output; evaluating the value of adopting a specific action in the current state; and processing the privilege information subjected to normalization processing by using a Cost network, and evaluating the cost of the current action of the quadruped robot.

The invention provides a foot type robot task planning control method and system, and relates to the technical field of robot control, and the method comprises the steps: obtaining a task instruction and multi-source sensing data of a robot; extracting the task instruction to obtain a semantic feature, and constructing a structured environment representation according to the multi-source sensing data; based on the semantic features, performing retrieval from a preset historical task library to obtain historical task experience fragments corresponding to the task instruction; a hierarchical task plan of the robot is generated through a language model according to historical task experience fragments in combination with structured environment representation, feasibility and safety verification is carried out, a target skill sequence of the robot is obtained, skill arrangement is carried out, a skill execution scheduling strategy is generated, and the robot is driven to execute the skill; and meanwhile, when the robot executes skills, real-time closed-loop correction is performed on a skill execution scheduling strategy according to real-time sensing data of the robot. According to the invention, the execution effect of the foot-type robot in a complex long-time task is improved.

The invention relates to a multi-modal data fusion-based multi-target tracking method for a legged robot. The method comprises the following steps of S1, data acquisition and preprocessing; s2, extracting data features; s3, multi-channel feature fusion extraction: based on the high-dimensional feature representation obtained in the S2, fusing the data features of the data of different sensor channels to obtain a multi-modal feature map, and processing the multi-modal feature map through a multi-task detection head to obtain a target detection result; s4, multi-target tracking: fusing data features of different sensor channels by using a feature sharing module, and outputting the fused data features to a multi-target tracking module; and the multi-target tracking module predicts the motion trail of the target in combination with the state of the target in the previous frame. According to the multi-modal data fusion-based multi-target tracking method of the legged robot, the accuracy and robustness of subsequent target detection and tracking can be improved, and continuous and stable identification and tracking of a sensitive target in a complex high-dynamic scene are realized.

The invention discloses a wheel-legged robot whole-body motion control system and method considering slippage steering, and relates to the technical field of wheel-legged machine control. The system comprises an input module used for receiving an expected motion track or a motion instruction from an upper layer planning system; the trajectory optimization module takes the motion trajectory output by the input module as a tracking target, so that the platform can be optimized in real time along an offline trajectory; the tracking control module is used for solving a joint torque through inverse dynamics by adopting a total rigid body dynamic model, realizing motion tracking through a whole body control WBC method, and solving a motion and contact force priority task by using a hierarchical optimization method; and the state estimation module is used for estimating all current state quantities of model predictive control MPC and information required by whole body control WBC through sensor information. By means of the system, high-dynamic stable motion control over the wheel-legged robot under the complex terrain can be achieved.

The invention discloses a bionic walking feature fused quadruped robot gait reinforcement learning training method, which comprises the steps of S1, bionic gait feature modeling for extracting key features from a motion mode of a natural quadruped animal and constructing a bionic template capable of directly guiding robot gait control; and S2, constructing a reinforcement learning training environment, and simulating diversified actual scenes by constructing a high-fidelity simulation platform. According to the method, key features are extracted from natural four-footed animal gaits, a bionic template library is constructed, and natural features such as nonlinear rhythm and dynamic symmetry of animal movement are fused into robot gait control, so that the problem of action mechanical stiffness caused by dependence on manual design of a track in a traditional method is effectively solved; movement of the robot is closer to natural biological gaits, impact generated when the robot interacts with the environment is reduced while movement energy consumption is reduced, and movement smoothness is improved.

The invention belongs to the technical field of legged robot control, and provides a quadruped robot robust motion control method and system based on joint learning, and the method comprises the steps: obtaining the body observation information and privileged observation information of a quadruped robot; extracting key features in the obtained privileged observation information by adopting a privileged encoder to obtain a first potential feature vector; obtaining a second potential feature vector matched with the first potential feature vector based on the obtained historical ontology observation information and an adaptive network; constructing a joint loss function according to the obtained first potential feature vector and the second potential feature vector, and performing strategy network updating training by taking the minimum joint loss function as a target; and robust motion control of the quadruped robot is completed according to the trained strategy network.

The invention discloses a quadruped vehicle sensing and decision-making system based on reinforcement learning. The quadruped vehicle sensing and decision-making system comprises a multi-mode sensing module, a dynamic fusion module, a reinforcement learning decision-making module and an execution module. The multi-modal sensing module is used for collecting multi-modal environment information including visual, tactile and inertial data; the dynamic fusion module carries out dynamic fusion on the multi-modal data by using a Transform-based feature extraction and time sequence alignment method to generate a unified environment representation; the reinforcement learning decision-making module is used for analyzing the multi-modal sensing result on the basis of a Proximal Policy Optimization (PPO) algorithm, and generating an optimized path planning and dynamic obstacle avoidance instruction; and the execution module generates a gait and motion control command according to the decision instruction to ensure the autonomous navigation and environment adaptability of the quadruped robot. According to the system, the sensing precision and the decision-making efficiency of the quadruped robot in a complex scene are remarkably improved, and the system has a wide application prospect.

The invention discloses an in-tunnel unmanned aerial vehicle and robot dog combined positioning method and system based on prism reference points, and belongs to the technical field of information processing. The method comprises the following steps: constructing a tunnel prism reference point array, acquiring an initial reflection signal, and preprocessing the initial reflection signal to obtain a processed signal set; determining a signal dynamic threshold, and obtaining a reference point set based on the signal dynamic threshold; acquiring reflected signal data of the unmanned aerial vehicle and the robot dog through the reference point set, and extracting reflected signal features; determining initial position distribution of the unmanned aerial vehicle and the robot dog based on the reflected signal features; obtaining dynamic reflection data of the unmanned aerial vehicle and the robot dog on the basis of the initial position distribution, and processing the dynamic reflection data to obtain a smooth position sequence; calculating a relative motion trend based on the smooth position sequence and motion difference characteristics of the unmanned aerial vehicle and the robot dog; the position consistency of the unmanned aerial vehicle and the robot dog is determined based on the relative movement trend; and determining real-time positioning coordinates of the unmanned aerial vehicle and the robot dog based on the position consistency.

Provided are a planning and control method for a legged robot, an apparatus, a robot, and a storage medium. The method includes: determining, based on the movement state and the swing parameter of each leg at the next moment, a foot-end reference state of each leg, and planning, based on a foot-end dynamic model and a discrete collision model, an impact-aware swing leg trajectory for a leg that starts to swing at the next moment; and performing, based on the movement state, the reference state sequence, and the foot-end reference state, whole-body control on the legged robot, and performing, based on a whole-body dynamic model and the discrete collision model, impact-aware whole-body control for a leg that is a supporting leg in planning but does not touch the ground.

The invention discloses a multi-robot area coverage path planning method based on a wheel-foot type platform, and belongs to the technical field of intelligent robot cooperative control. According to the method, the characteristic that the wheel-foot robot can adapt to various terrains is combined, the environment is divided into a plurality of sub-regions through a multi-robot task pre-distribution method introducing terrain parameters, each robot is responsible for covering one sub-region, and the calculated amount and the search space of a path planning algorithm are reduced; in the sub-regions, a path covering all task regions can be constructed from any free grid through an improved STC algorithm, and 100% path coverage is realized under the condition of a grid map; in the actual task stage, the change of the actual terrain and the actual state is fed back in real time through the sensing ability of the wheel-foot robot to the environment, and task allocation and path planning are adjusted in real time.

The invention discloses a double-foot wheel-legged robot pose control method and device based on hierarchical cooperative control, a medium and a product, and relates to the field of robot control, and the method comprises the steps that a hierarchical cooperative control framework is constructed; the layered cooperative control framework comprises a balance controller and a whole body attitude controller; the balance controller determines a reference wheel joint driving torque based on a simplified inverted pendulum dynamic model according to the pose information of the upper mass center of the biped wheel-leg robot; the whole-body attitude controller performs optimization control on all joints of the biped wheel-leg robot by adopting an incremental model prediction control algorithm combined with a reference wheel joint driving moment based on a whole-body dynamic model; state information of the biped wheel-legged robot is obtained; according to the state information, a driving torque instruction of each joint is obtained based on a hierarchical cooperative control framework; and driving the double-foot wheel-leg robot to move according to the driving torque instruction of each joint. The precision, robustness and real-time performance of robot motion control can be improved.

The invention relates to a hexapod robot leg impedance control method and device, equipment and a medium, and relates to the technical field of automation control, and the method comprises the steps: training a primary function neural network through domain adversarial invariant learning based on a force-displacement compensation data set; a robot contact force deviation value is calculated, the contact force deviation value is input into the primary function neural network to extract environment invariant features, a Kalman filter is utilized to estimate an adaptive coefficient on line according to the environment invariant features, and a displacement compensation amount is calculated in combination with the environment invariant features and the adaptive coefficient; and the displacement compensation amount is injected into a robot impedance control loop, and leg movement of the hexapod robot is adjusted. According to the method, the environment invariant features are extracted, so that the method can better adapt to various complex terrains, the sensitivity to environment parameters is reduced, and the urgent requirements of the hexapod robot on high robustness, high precision and rapid self-adaptive contact control in actual operation are met.

The invention provides an interaction control method and device of a foot type robot, equipment and a medium. The interaction control method comprises the following steps: determining local accumulated point cloud data according to pose transformation information and registered point cloud data; performing passable judgment on the foot-type robot through the local accumulated point cloud data to obtain a local passable region; according to the local accumulated point cloud data, low obstacle point cloud data and impassable obstacle point cloud data of the foot-type robot in a local passable region are determined; clustering the low obstacle point cloud data to obtain a low obstacle representation shape; clustering the impassable obstacle point cloud data to obtain an impassable obstacle representation shape; and rendering the low obstacle representation shape and the impassable obstacle representation shape into a local passable region displayed in a first display interface at the handle end of the foot robot in real time. Therefore, the user experience is effectively improved.

A system in accordance with at least some embodiments of the present technology includes a robot and a dock. The robot includes a body and a plurality of legs connected to the body through which the robot is configured to ambulate. The robot further includes a hanger carried by the body and a charge-receiving electrode at the hanger. The dock includes a hook, a charge-dispensing electrode at the hook, and a guide that urges the hanger into alignment with the hook. The system is transitionable between an undocked state and a docked state. In the undocked state, the robot and the dock are spaced apart from one another. In the docked state, the hanger is received at the hook, the dock supports at least a portion of a weight of the robot via the hook, and the charge-receiving electrode is electrically connected to the charge-dispensing electrode.

The present invention discloses a novel closed-chain octapod robot, belonging to the field of closed-chain legged robots; it includes a frame, four walking leg assemblies, and a driving mechanism; the walking leg assembly includes a gear transmission mechanism, a deformable octapod leg mechanism, and several connecting pieces, the gear transmission mechanism includes a first gear set and a second gear set, the driving mechanism includes an adjustment servo installed on the walking leg assembly, and also includes a driving motor installed on the frame and a one-to-four belt transmission mechanism; the driving motor and the adjustment servo drive the octapod leg mechanism to swing back and forth, and the four groups of octapod leg mechanisms coordinate with each other to enable the octapod robot mechanism to perform walking actions. The present invention selects a closed-chain octapod structure as the walking mechanism of the legged robot. This solution has the advantages of few driving numbers, low energy consumption, large overall stiffness, and strong load capacity, and is an important research field for the innovation of the leg configuration of bionic robots.

The invention relates to a bimorphic self-adaptive driving device of a legged robot, and belongs to the technical field of driving devices.The bimorphic self-adaptive driving device comprises a bottom plate, a connecting disc used for mounting the robot is fixedly mounted at the top of the bottom plate, a jacking mechanism used for jacking the bottom plate is arranged in the bottom plate, and mounting frames are symmetrically arranged at the bottom of the bottom plate; a connecting frame is fixedly mounted at the bottom of the mounting frame, a concave frame is arranged in the mounting frame, and a turnover mechanism used for adjusting the angle of the concave frame is arranged in the connecting frame; when walking on a flat road surface, the robot can move in any direction in a plane through the assembled Mecanum wheels, when the road surface condition is bad, the turnover mechanism turns over and turns into another mode, the five-section type module driving design is adopted to simulate the movement of a snake, each section is provided with an independent driving device and an independent control system, and therefore the driving efficiency of the snake is greatly improved. The robot can flexibly control the motion posture of each module, so that different motion modes are realized, and the flexibility is improved.

The invention relates to a quadruped robot parkour navigation method and system based on multi-modal feature fusion, and the method comprises the steps: constructing a multi-modal parkour topographic data set which comprises a depth image, body feeling information and a control instruction; encoding the depth image sequence and the body sensing sequence of the historical time step to obtain space-time fusion features; encoding the control instruction and the corresponding target position vector to obtain a historical decision feature; encoding a depth image collected at the current moment to obtain fine spatial features; fusing the space-time fusion feature, the historical decision feature and the fine space feature to form a conditional context; and based on the diffusion strategy model, by taking the conditional context as a condition, carrying out de-noising processing, and generating a navigation control instruction so as to complete a parkour navigation task. Compared with the prior art, the method has the advantages that the navigation success rate is high, path planning is diversified, complex 3D terrains can be effectively processed, and the problem that high-level planning and bottom-level motion control are disjointed is solved.

The invention discloses a fall and turn-over control method and device of a foot type robot and a storage medium. The fall and turn-over control method comprises the steps of firstly obtaining fall posture information of the foot type robot; calling an action selection model to perform multi-stage iterative turnover action prediction based on the falling posture information to obtain an optimal turnover action of each iteration stage, controlling the foot type robot to execute the optimal turnover action of the current iteration stage in each iteration stage, and controlling the foot type robot to execute the optimal turnover action of the current iteration stage after the foot type robot executes each optimal turnover action. The foot-type robot is in a stable posture when the foot-type robot turns over, and the foot-type robot turns over successfully after executing the optimal turning-over action of the last iteration stage. Wherein when the action selection model is called to carry out multi-stage iterative turning-over action prediction based on the falling posture information, the action selection model predicts the optimal turning-over action of the next iteration stage according to the stable posture of the foot robot in the current iteration stage. According to the embodiment, the manual design cost can be reduced, and the applicability is high.

The invention discloses a decoupling motion control method and system for a four-footed mobile operation robot considering the acting force of a mechanical arm, and belongs to the field of motion control of foot type mobile operation robots. An expected motion track input by a user can be tracked by performing planning through linear model prediction control MPC; on the basis of the tracked motion trail, mechanical arm joint control torque is calculated through a mechanical arm dynamic model and PD feedback; through nonlinear model predictive control NMPC planning, a quadruped robot whole-body motion trail of mechanical arm acting force obtained through calculation according to a mechanical arm dynamic model is considered, and an expected speed trail and an expected force trail input by a tracking user are obtained; and according to an expected speed trajectory and an expected force trajectory input by a tracking user, a whole body controller WBC based on hierarchical quadratic programming calculates a joint driving torque for tracking the expected trajectory according to task priorities. According to the method, the acting force / torque of the mechanical arm to the robot body is considered during motion control of the quadruped robot, and decoupling control over the mechanical arm and the quadruped robot in the quadruped mobile operation robot is achieved.

The invention discloses a quadruped robot control method and system based on a Transform reinforcement learning architecture. According to the method, the body state and environment information of the quadruped robot are obtained in real time through a multi-mode sensor; constructing the information of the current moment and the historical moment into a time sequence state sequence; performing feature embedding and time position coding on the sequence to form an input token sequence; inputting the token sequence into a strategy network which takes a Transform encoder as a core and is pre-trained through reinforcement learning; the network captures a long-range global dependency relationship between states by using a self-attention mechanism, and outputs a continuous action instruction of a joint level; and finally, each joint of the robot is driven by the bottom layer controller to execute corresponding motion. The system comprises corresponding sensing, calculating and executing modules. According to the method, the problems of gradient disappearance and memory attenuation of a traditional recurrent neural network in long sequence decision making are effectively solved, and the motion intelligence, robustness and energy efficiency of the quadruped robot in a complex and unstructured environment are remarkably improved.

The application discloses a novel mobile grabbing integrated bionic multi-legged robot, which comprises a front body provided with two leg parts and a rear body provided with four leg parts; the front body and the rear body are connected through a tension structure capable of being lifted and deformed; when a driving wire is driven to be tightened, the tension structure can be lifted and deformed upward; when the driving wire is loosened, the tension structure can be lowered and restored under the action of a reset spring; the front body and the tension structure move synchronously, so that the front body can be lifted and deformed or lowered and restored. When the front body is not lifted, the robot moves in a three-legged gait through six leg part mechanisms; when the front body is lifted, two leg parts of the front body serve as a mechanical arm, a steering wheel at a hip joint is used to realize a clamping function, the robot moves through four full leg parts of the rear body, and a four-legged gait is adopted, so that the robot can complete a carrying operation.

The invention discloses a wheel-legged robot for teenager education and a control system thereof.The wheel-legged robot comprises a vehicle-mounted frame and a driving assembly arranged on the outer wall of the vehicle-mounted frame, and the driving assembly comprises a joint motor, a driving arm, a torsion frame, a torsion spring, a first lifting rod, a second lifting rod, a driving wheel, a limiting shaft, a driving motor and a control assembly; the joint motors are fixedly arranged on the two sides of the outer wall of the vehicle-mounted frame. A four-connecting-rod and torsion spring balance mechanism is used as a core, smooth wheel-leg switching, easy obstacle crossing, LQR and reinforcement learning fusion, online self-adaption, more stable and more power-saving movement and voice, vision and gesture multi-mode interaction facing teenagers are achieved, complex tasks can be completed by speaking sentences and moving hands, a low-cost motor is developed by itself, a full-open-source design is adopted, and the robot is more convenient to use. Students can disassemble, assemble, expand and re-create, mechanical, electronic and programming knowledge is communicated in practical practice, and the education of department and creations is really extended from class to family and future.

The invention relates to a quadruped robot motion control method based on visual fault tolerance, and the method comprises the steps: building a simulation environment according to the motion control demands of a robot; defining a blind estimator and a visual estimator, organizing a training environment according to scene requirements, and training a switching strategy between the blind estimator and the visual estimator; determining the motion constraint of the robot based on a cMDP constraint Markov decision process, and solving a robot motion control strategy in combination with the trained blind estimator and visual estimator; and correspondingly controlling the motion state of the robot according to the robot motion control strategy. Compared with the prior art, the advantages of visual estimation and ontology perception are combined, the robot can move more flexibly and stably in the complex terrain, a switching strategy between estimators is designed, the visual fault-tolerant capability can be enhanced, the motion stability of the robot is guaranteed, in addition, an end-to-end one-stage training framework is adopted, and the training efficiency is improved. The training and deployment efficiency can be improved.

The utility model discloses a leg device of a foot type robot and the foot type robot. The leg device of the foot type robot comprises a robot leg and further comprises a driving motor arranged in the robot leg, the driving motor comprises a shell, a power cavity formed in the shell and a power assembly arranged in the power cavity, and a first containing cavity is formed in the end face of the tail of the shell; a motor controller electrically connected with the power assembly is arranged in the first containing cavity, and cooling fins are arranged on the motor controller and connected with the shell in a heat conduction mode. The utility model further discloses the foot type robot which adopts the leg device of the foot type robot. The motor controller has the advantage of solving the problem that the motor controller for installing the motor in the leg of the humanoid robot is easy to overheat. The cooling fins are connected with the shell in a heat conduction mode, heat generated by the motor controller can be rapidly transmitted to the shell and then dissipated to the surrounding environment through the shell, faults caused by overheating of the motor controller are avoided, and stable operation of the motor controller is guaranteed.

The invention discloses a chain type hybrid deep reinforcement learning quadruped robot control system. The system comprises a sensing module, a data processing module, a chain type hybrid reinforcement learning model module, a feedback module and the like. The invention further discloses a control method of the chain type hybrid deep reinforcement learning quadruped robot control system, and the method comprises the steps: S1, collecting bionic motion data, robot state data and environment data, and training based on an attention mechanism space-time diagram neural network and antagonistic imitation learning to obtain a reference gait model containing motion prior; and S2, constructing a three-level parallel simulation environment, setting a progressive course target, training the progressive course target by adopting an improved PPO algorithm, and optimizing the gait stability, the movement efficiency and the environmental adaptability. By means of high-precision data acquisition with the sampling frequency larger than or equal to 240 Hz and antagonistic imitation learning, motion priori knowledge is introduced, the training period can be shortened by 50% or above, abnormal gaits are avoided, and the training process is controllable.