Implement a custom replay buffer for octree observations

AndrejOrsula / drl_grasping

Deep Reinforcement Learning for Robotic Grasping from Octrees

https://arxiv.org/pdf/2208.00818

BSD 3-Clause "New" or "Revised" License

404 stars 54 forks source link

Implement a custom replay buffer for octree observations #61

Closed AndrejOrsula closed 3 years ago

AndrejOrsula commented 3 years ago

Two main things that this replay buffer should address:

Create octree batch directly (regular batch is incompatible)
- This would reduce the number of RAM -> VRAM copies
Allow variable-sized observations instead of 0-padding

AndrejOrsula commented 3 years ago

First part (Create octree batch directly) implemented in https://github.com/AndrejOrsula/drl_grasping/commit/363d21c9fb7eee070489fd990cdd869ecb14f618 and offers a minor performance boost

I am skipping the second part (Allow variable-sized observations instead of 0-padding) as it would most likely have significant negative impact on the performance.