Page 107 - Read Online

Page 294 Zhang et al. Intell Robot 2022;2(3):275­97 I http://dx.doi.org/10.20517/ir.2022.20

https://github.com/bulletphysics/bullet3/tree/master/examples/pybullet/gym/pybullet_envs/minitaur/envs

/ / / / / / / https://github.com/OpenQuadruped/spot_mini_mini / / / /
Latency), Iner- and bias Additive Bodies State. Foot Masses, Randomisa- Size, and

Model, Strength, Motor IMU Model, Randomised Initial and Link Domain Mass Link

(Actuator (Mass, Friction, Forces. Actuator Training, Command Leg , Mass and Scaling, Time).

Fidelity Randomization Battery, Latency, Virtual Properties, Curriculum Random (Base Magnitude). Actions, and Torque Step

Simulation Step, Directional Physical Observation. Model, Position), Randomization Mesh XYZ Observations Actuator and Damping,

Improving Dynamics and Control tia, noise. Random Randomized the to Noise / / / / tra- Actuator and (Size Domain Gait Friction, Noisy (Gravity, tion Actuator / / Si- /

Controller. Generator. Generators. Recorded Data (Joint Angles and steps). 4800 for foot generating actuator the train curve Bezier model-based Planning), (High-Level Low-Level (TG, Policy

Open-loop Trajectory / Trajectory PMTG [49] Orientation CROC [110] controller A to jectories model. Open-loop Generator. / simple A method [45] . PMTG MPC nusoidal Controller).

Agile Quadruped Trajectory Deep Robots [99] Quadruped Quadruped through Primi- Con- us- Gaits Learn- mo- Legged Policy Dynamic Locomo- Efficient Generaliza- Locomotion Locomo- Re-

Learning For Modulating RobustRecovery Controller fora using Robot Learning [48] Reinforcement Legged Reinforcement for Learned Behaviors Motion and Planning Quadrupedal Reinforcement dynamic and robots [6] legged DynamicsandDomainRandom- izedGaitModulationwithBezier Sim-to-Real Constrained for Robot Contact-Adaptive Robust, Locomotion [101] Terrain Visual Generalizable Hierarchical Learning [47]

Sim-to-Real: Locomotion Robots [39] Policies Generators [49] Quadrupedal Reinforcement Efficient Data for Learning Hierarchical Learning Locomotion [46] Realizing Locomotion Kinematic tives [55] DeepGait: of trol Deep ing ing [100] agile Learning for skills tor for Curves Locomotion [50] Guided Optimization Quadrupedal tion [28] a Learning for Controller Legged Zero-Shot for tion Policies [103] Learning with Skills tion inforcement