Tom Lenaerts, along with The Anh Hanh, Francisco C. Santos, and Luís Moniz Pereira, published Voluntary safety commitments provide an escape from over-regulation in AI development. The journal paper, published in Technology and Society, shows that voluntary, yet enforced, commitments lead to safe development behaviour and avoiding problems of over-regulation. 

Arnaud Fickinger and Stuart Russell, along with Brandon Amos and Samuel Cohen, published Cross-Domain Imitation Learning Via Optimal Transport. Included in the 2022 International Conference on Learning Representations, the paper provably achieves cross-domain transfer in non-trivial continuous control domains by minimizing the Gromov-Wasserstein distance with deep reinforcement learning.

Scott Emmons, along with Ben Eysenbach, Ilya Kostrikov and Sergey Levine, published RvS: What is Essential for Offline RL via Supervised Learning? Included in the 2022 International Conference on Learning Representations, this work finds that surprisingly simple design choices are sufficient for strong empirical performance on offline RL benchmarks. It also shows that using supervised learning to learn a goal-conditioned policy — one that can condition on arbitrary goal or reward targets at inference time — is competitive with offline TD learning methods.