To run this project, you should first clone the repository: https://github.com/ANYbotics/anymal_b_simple_description.
Put the files in the root path of this repository, start MATLAB, and open 'Anymal_B_example.prj' to check if you can load the project successfully.
To train a DDPG or TD3 agent, you can use 'testDDPGAgent.mlx' and 'testTD3Agent.mlx' in MATLAB, set 'doTraining = false' to 'doTraining = true' (line 40), if necessary. These scripts utilizes the files 'utility\createDDPGAgent.m' and 'utility\createTD3Agent.m' to create the DDPG and TD3 agent respectively, and share the same file 'utility\createNetworks' to create actor/critic networks.
To change the device for training and inference, you can alter the 'UseDevice' value, in criticOptions and actorOptions within 'utility\createDDPGAgent.m' and 'utility\createTD3Agent.m',
Check the 'rlQuadruped_TD3.avi' to see how the TD3 agent performs, for more information, such as the trained TD3 agent and reinforcement learning workflow in MATLAB & Simulink, please contact MathWorks directly.