Page 109 - Read Online

Zhang et al. Intell Robot 2022;2(3):275­97 I http://dx.doi.org/10.20517/ir.2022.20 Page 295

https://github.com/erwincoumans/motion_imitation

/ / / / https://github.com/ZJU-XMech/PhaseGuidedControl / / / / /
Teach- Torque, Noise of Joint Velocity). and Strength Friction, and Height Mass Base De- and Strength friction).

Adapation. Parameters, Restitution). and Force Angular and Motor Joint Battery, Step and Mass, and Communication Motor Joint

Domain Physical StudentTraining Set-up, and Automated Curriculum synthe- Loss [112] . Domain randomization (Link Mass, Inertia and CoM, Initial and (External Posture, Inertia, Friction, force, Profile and Inertia, Battery,

and Randomized Smoothing Friction Ground Friction and Height, Mass, Body Size, Body Identification. (Mass, Lateral External (Leg Matrix, Inertia (Mass, friction,
Randomization Model, Terrains. and Filtering Ground and Randomization Velocity, and Simulator Randomization Latency, noise, position Randomization Leg Randomization Lateral Latency,

Domain Actuator sizing Action Direction / Domain Position / Hybrid Domain Friction, CoM width. Domain Distribution, lay). Domain Friction,

Joint Con- Inverted Dynamics

Generator. provide Foot and Linear (Model-based

Clips [111] . Trajectory to Gait Reward Spring Hopf Oscillator and manually de- functions [114] . Control. Generator.

MoCap Foot Reference Position Reward. tact SLIP: Pendulum Controller). TOWR [113] signed Centroidal Model [115] . / Optimal Trajectory PMTG [49] .

Loco- Ani- Locomo- Terrain [7] Control Quadruped Neu- Terrain- Imitat- Plan- Transition Locomotion Simulator Domain Adversarial Locomo- Rein- for Based Re- for in

Robotic Imitating of Pretrained by Dynamics Gait Transitions [52] for Learning [104] Deep Framework Locomotion Learning [105] Risk-Assessment- Deep Learning Locomotion

Agile by Skills Quadrupedal Challenging Multi-expertlearningofadaptive locomotion [9] Learning Robust for using Networks [57] Coordinated Locomotion Centroidal Free for Quadruped Robots via Phase- Controller [58] Efficient Gait Hybrid via Hierarchical Terrain-Aware Con- Quadrupedal Combining forcementLearningandOptimal Hierarchical Reinforcement Terrain [51]

Learning motion mals [10] Learning over tion legged Efficient Policies Bounding ral Learning Adaptive a ing ner [27] Learning Guided and Fast Learned via SimGAN: Identification Adaptation Reinforcement for trol by tion Control [54] A Quadruped on Terrain-Aware Network-Aided inforcement Quadrupedal Tough