Page 111 - Read Online
P. 111

Page 296                                Zhang et al. Intell Robot 2022;2(3):275­97   I http://dx.doi.org/10.20517/ir.2022.20





























                                                                                                https://github.com/lauramsmith/fine-tuning-locomotion


                                                                                                                            https://github.com/leggedrobotics/legged_gym




















                               /         /        /       /      /        /          /                      /        /
                            Ini-  Scaling,                     Coef-  and  Velocity,  Fric-             Scaling,  on  Ac-  External
                            Shifting  Torque                   Friction  Gain,  Angular  Ground         Torque  Perturbation  Map,  and


                            Modelling,  Actuator               Ground  Derivative  and  Velocity,       Actuator  Damping),  Elevation  for  Friction




                            Actuator  Gravity,  Base.  Module [60] .  Module.  Module.  Mass,  (Link  and Proportional  Linear  (Base  and  Position  Disturbances).  (Gravity,  Actuator  Filters  (Ground  Observations.


                            Randomization,  Changing  Robot  the  Adaption  Adaption  Adaption  randomization  Slope,  Delay).  randomization  Joint  Gravity,  External  Fine-Tuning.  randomization  Size,  and  Smoothing  and  Randomization.  Randomization  Noisy






                            Domain  Position,  tial  Perturbing  and  RMA-based  RMA-based  RMA-based  Domain  and  ficients  Communication  Domain  Projected  and  tion  /  Real-World  Domain  Mass  Link  Base,  Robot  Model.  tuator  Domain  Domain  and  Force),




                                                                                   Algo-          A1 [10] .


                               WBC [116] .  librealsense 4 .     Motions.          Optimization  Recording [111] and  for  motion  Controller.  Strategy.



                               TOWR [113] ,  FMM [117] ,  /  /   Human    /        Trajectory  rithm [118] .  Dog  MoCap  Side-Step  Gaits  Dynamic  Curriculum  /






                           Adapta-  Locomo-  Reinforcement  Legged  Consump-  of  Emergence  Adaptation  of  Control  for  Control  Reinforcement  Locomo-  from  on  Keep  Loco-  Real  the  Legged  Reinforce-  Optimal  Us-  Minutes  4 https://github.com/IntelRealSense/librealsense



                           Trajectory  Quadrupedal  Deep  Coupling Vision and Propriocep-  of  navigation  Energy  the  to  Robots [107]  Legged  Motor  Robots [60]  Motion Quadrupedal Robots using Deep  Learning [108]  Torque  Locomotion [44]  Robust  for  Demonstrations  Optimization [109]  that  Robots  Fine-Tuning  in  Policies  Terrain-Aware  using  and  Learning  RapidLocomotionviaReinforce-  Learning [36]  in  Walk ing Massively Parallel Deep Rein-  Learning [35]



                           Real-Time  for  tion  using  tion  Learning [53]  for  tion  Robots [11]  Minimizing  Leads  tion  in  Gaits  Rapid  RMA:  Legged  for  Human  Reinforcement  Learning  Quadrupedal  Model-free  Learning  using  tion  Trajectory  Legged  Learning:  motion  World [30]  RLOC:  Locomotion  ment  Control [42]  ment  to  Learning  forcement
   106   107   108   109   110   111   112   113   114   115   116