Can we use depth cameras as tactile sensors for quadruped locomotion?
Reference
This work is built on top of Rex-Gym: https://github.com/nicrusso7/rex-gym
What code is written by me?
Added render_necek function and render_foot function in rex_gym/envs/rex_gym_env.py to add depth cameras to mimic tactile sensor.
Created a new robot rex_gym/util/pybullet_data/robot_tactile.urdf which has shell on feet to hold cameras
Modified rex_gym/envs/gym/walk_env.py to feed depth image observation to the training pipeline.
Modified the global parameters in rex_gym/model/terrain.py to generate different terrains for experiments.
Modifed the network structure in rex_gym/agents/scripts/networks.py by adding new observation layers.
Modified rex_gym/playground/policy_player.py so I note the status of robot.
Created main.py so I don't need to install the package.
Beyond writing code
Most importantly, to make these modifications and to make them work, I read and understood ALL code in this repository and learned how to use RNN in tensorflow(which is much unfridenly to beginners than pytorch in my opinion).