Untitled

This paper:

• extends periodic learning method [4] by including reference traj. data, which is derived offline using a single rigid body model (SRBM)
• simply applies proposed learning method to sim-to-real
• switches between policies trained for each desired behavior
• performs highly dynamic behaviors such as high speed turning

Contents:

• Previous Works
• Methods
• Results

Previous Works

1. RL-based methods [4], [7], [9]:

• To perform different behaviors, use RL to train a single policy for all desired behaviors

2. Model-based methods [3], [13]-[15]: