Page 103 - Read Online

Page 292 Zhang et al. Intell Robot 2022;2(3):275­97 I http://dx.doi.org/10.20517/ir.2022.20

A1 A1 A1 A1 A1 A1 A1
Yobogo Unitree ANYmal Unitree Unitree Unitree Unitree Unitree Solo8 Unitree

Webots [106] Foot Pybullet RaiSim Tra- RaiSim RaiSim RaiSim RaiSim and Ori- Foot Pybullt Isaac, Pybullet End- Pybullet Ve-

BaseTargetPositions(withPrevi- Survival. and Orientation, Velocity, Angular and Collision, and Motion Smoothness, Target Clearance, Map. Traversability and Smoothness, State, Foot Motion, Joint Orientation, Goal. Tracking, Consump- Energy Gap, and Hip Lateral Movement, Linear and Angular Velocity, and Velocities. and Torques Orientation, and Motion Smooth- Impact, Ground Slip. Foot Speed, Joint End-Effector, Imitation, Deviation, Sate, Acceleration. Height, and Velocities Motion, Joint Action Collision, Knee Reward. Hips and Gaits, Position, Base and Position, Veloc- Angular

Base and ous), Height. Linear Base Joint Torque, Ve- Torque, End- Ve- and jectory Velocity Joint tion, Joints. Joint Joint Base Joint Work, ness, Joint Joint Base Base entation, State, Rate, Joint Joint ity. Joint Joint Effector locity.

Position, Pitch, Max. and State Deviation, State Deviation. Torques.

Goal Velocity, Lift Leg Desired Positions. Base locity Effector locities Desired Positions. Desired Positions. Desired Positions. Desired Positions [43] Desired Desired Positions. Desired Positions.

and Direction, and Frequen- His- Con- Env. Velocities, Veloc- Po- Height Ve- Ex- and Velocities, Velocities, Foot Ac- Reference Po- Joint Command, Velocities, and and An- (3-step),
Orientation, Goal State Joint Phases, Joint FTG), and Targets States, Contact and and Positions Trajectory Velocities, Command Action and Orientation, Foot Contact, Previ- and and (2-dim), history), (with and history). Gravity, Velocity, and Roll) and (Yall Joint (3-step), Actions Poses.

Error, Map, Velocities, FTG and (Base Foot Forces, Positions Corrected and Proprioception, Previous Vector. Positions Action. Positions Vector. Data History, (with Velocities, and Action. Positions (Yall Velocities (3-step), Target

Angle Command. Elevation Base Velocity, cies tory, tact Param. Robot Reference ities, sitions Maps. locities, trinsic Joint ous Joint Orientation Contact Sensor tions Poses Base sition Last Joint Orientation Angular Roll). Orientation gles Future

REDQ [29]
SAC SAC PPO PPO PPO PPO PPO PPO PPO

quadruped lo- learning terrains con- DRL-based stability. noisy get to to order in tracking propri- and of tasks locomo- to related is real-time for in problems control motion opera- human control torque high- via torques robust create us- training demon- fine- for locomotion

well-performing for system real-world in pre-training. terrain-aware troller integrating a RAN to guar- action the DRL using trajectory quadrupedal a vision Incorporating navigation in robots. Energy constraints leading to the natural of choice the and speed. desired algorithm adaptation robots. quadrupedal allowing quadrupedal predicting joint to approach policiesdeployableonrealrobots additional optimised single system RL real-world

A robot comotion without A antee policy A reference generate system. oception legged emergence tion, the RMA online quadruped A system tion. A framework frequency RL. RL A without a ing stration. robot A tuning policies.

2021 2021 2021 2021 2021 2022 2022 2022 2022
IROS IROS ICRA CVPR workshop 2021 CoRL RSS RSS ArXiv ArXiv ICRA

for on for Tough Adaptation using in Gaits for of DRL [108] for Learn- Policies

Framework Based Locomotion Risk-Assessment- DRL (RAN) in Locomotion Trajectory Locomotion Propriorception and Robots [11] Legged Consumption of Emergence Adaptation Control using Control Locomotion [44] Model-freeRLforRobustLocomotion usingDemonstrationsfromTrajectory on Keep Locomotion

Hierarchical Terrain-Aware Network-Aided Quadrupedal Quadrupedal Vision of navigation Energy the to Robots [107] Motor Rapid Robots [60] Motion Robots Quadrupedal Torque Quadrupedal Optimization [109] that Robots Fine-Tuning World [30] Real

A Quadruped RL [105] Terrain [51] Real-Time for DRL [53] Coupling for Minimising Leads Legged RMA: Legged Human Learning Legged ing: the in