Deep Mimic with BVH Data

Code accompanying the SIGGRAPH 2018 paper: “DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills”. The framework uses reinforcement learning to train a simulated humanoid to imitate a variety of motion skills from mocap data. BVH Biovision file format. .bvh files only contain the final skeleton and skeletal animation. We find BVH files in ASCII format easy to use in our own programs.

Code: https://github.com/bsivanantham/DeepMimic

Network Architecture

• Policy π is represented by a neural network that maps a given state s and goal g to a distribution over action π(a|s,g).
• The inputs are processed by two fully-connected layers with 1024, and 512 units each, followed by a linear output layer. ReLU activations are used for all hidden units.
• Permalink:https://github.com/bsivanantham/DeepMimic/blob/3c7f2171bb09512bee47d667de39761cde9b2d53/learning/nets/fc_2layers_1024units.py#L6

Data