Simple Emergent Action Representations from Multi-Task Policy Training, is accepted to Deep RL Workshop@NeurIPS 2022! arXiv, Website