Open Issues Need Help
View All on GitHub Find out which of the eight inputs the quadruped actually uses about 2 months ago
good first issue
Bug: viewer --random steals the environment's random number generator about 2 months ago
bug good first issue
Bug: a constant action beats the trained policy about 2 months ago
bug good first issue
Make the quadruped stand still and balance instead of walking about 2 months ago
good first issue
Make the quadruped walk as fast as you can about 2 months ago
good first issue
Make the quadruped walk backward about 2 months ago
good first issue
Bug: the viewer draws left and right feet on top of each other about 2 months ago
bug good first issue
Bug: a leg can never actually lift off the ground about 2 months ago
bug good first issue
Bug: the legs never form a real gait — the pattern is whatever reset() randomised about 2 months ago
bug good first issue
Bug: the quadruped can never fall over, so the fall penalty does nothing about 2 months ago
bug good first issue
Add running observation normalization about 2 months ago
good first issue
Trim per-step and eval overhead in the training loop about 2 months ago
good first issue
Leg-topology-aware (graph/symmetry) architecture instead of a flat MLP about 2 months ago
enhancement good first issue
Add return logging/plotting so training progress is measurable about 2 months ago
good first issue
Trim per-step and eval overhead in the training loop about 2 months ago
good first issue
Add return logging/plotting so training progress is measurable about 2 months ago
good first issue
Bootstrap behavior for the cold-start (no tactile history yet) problem about 2 months ago
good first issue
Leg-topology-aware (graph/symmetry) architecture instead of a flat MLP about 2 months ago
enhancement good first issue