There’s only 8 params for the biped walker, compared to thousands of params for the policy network. In this case it should be possible to scale to the thousands

There’s only 8 params for the biped walker, compared to thousands of params for the policy network. In this case it should be possible to scale to the thousands of environment parameters. For instance if we want to learn the material configuration of each voxel of a soft body robot assembled from thousands of voxels.