Miniprojects for the MICRO-507 : Legged Robots course
Project 1 is about planning the Center of Mass (CoM) trajectory for a (simulated) biped Atlas robot. The technique used is the Divergent Component of Motion (DCM). The resulting gaits are analyzed and discussed, and the conclusion is reached that this method requires quite a few simplifications, some of which are not valid in the real world.
Project 2 adresses quadruped locomotion. Gaits are generated for (simulated) A1 quadruped, first using Central Pattern Generators (CPG), and the using Deep Reinforcement Learning (DRL) The resulting gaits are analyzed and discussed, and the conclusion is reached that CPG methods are relatively simple and quick to implement, however, they require a lot of parameter tuning to get them to work nicely. DRL techniques require no (manual) parameter tuning, but instead require the setting an appropriate reward function, which can be difficult to guesstimate.
Folder Name | Comment |
---|---|
Project_1 | folder containing the project 1 |
- code | python code for the project |
Project_2 | folder containing the project 2 |
- code | python code for the project |
fast_locomotion.mp4
normal_locomotion.mp4
PACE.mov
WALK.mov
BOUND.mov
TROT.mov
RL_PPO_WITH_CARTESIAN_PD.mov
RL_PPO_WITHOUT_CARTESIAN_PD.mov
Biselx Michael - [email protected]
Bumann Samuel - [email protected]
lvuilleu
Project Link: https://github.com/mbiselx/LeggedRobots