Control and decision systems are becoming a ubiquitous part of our daily lives, ranging from serving advertisements or recommendations on the internet to controlling autonomous physical systems such as industrial equipment or robots. While these systems have shown the potential for significantly improving quality of life and industrial efficiency, the impact of the decisions made by these systems can also cause significant damages. For example, an online retailer recommending dangerous products to children, a social media platform serving content which polarizes society, or a household robot/autonomous car which collides with surrounding humans can all cause significant direct harm to society. These undesirable behaviors not only can be dangerous, but also lead to significant inefficiencies when deploying learning-based agents in the real world. This motivates developing algorithms for learning-based control which can reason about uncertainty and constraints in the environment to explicitly avoid undesirable behaviors. We believe hosting a discussion on safety in learning-based control at NeurIPS 2021 would have far-reaching societal impacts by connecting researchers from a variety of disciplines including machine learning, control theory, AI safety, operations research, robotics, and formal methods.
| Introduction (Short Talk) | |
| Ye Pu (Invited Talk) | |
| Ye Pu (Talk Q/A Session) | |
| Aleksandra Faust (Invited Talk) | |
| Aleksandra Faust (Talk Q/A Session) | |
| Shie Mannor (Invited Talk) | |
| Shie Mannor (Talk Q/A Session) | |
| Learning Contraction Policies from Offline Data (Oral) | |
| Safety-guaranteed trajectory planning and control based on GP estimation for unmanned surface vessels (Oral) | |
| Efficiently Improving the Robustness of RL Agents against Strongest Adversaries (Oral) | |
| Safe RL Panel Discussion (Discussion Panel) | |
| Poster Session I | |
| Rohin Shah (Invited Talk) | |
| Rohin Shah (Talk Q/A Session) | |
| Angelique Taylor (Invited Talk) | |
| Angelique Taylor (Talk Q/A Session) | |
| Ugo Rosolia (Invited Talk) | |
| Ugo Rosolia (Talk Q/A Session) | |
| Safe RL Debate (Debate) | |
| Reinforcement Learning with Feedback from Multiple Humans with Diverse Skills (Oral) | |
| What Would the Expert do()?: Causal Imitation Learning (Oral) | |
| Who Is the Strongest Enemy? Towards Optimal and Efficient Evasion Attacks in Deep RL (Oral) | |
| Poster Session II | |
| Closing Remarks (Short Talk) | |
| Distributionally robust chance constrained programs using maximum mean discrepancy (Poster) | |
| Robust Reinforcement Learning for Shifting Dynamics During Deployment (Poster) | |
| What Would the Expert do()?: Causal Imitation Learning (Poster) | |
| Who Is the Strongest Enemy? Towards Optimal and Efficient Evasion Attacks in Deep RL (Poster) | |
| Safe Learning of Linear Time-Invariant Systems (Poster) | |
| Efficiently Improving the Robustness of RL Agents against Strongest Adversaries (Poster) | |
| Safety-guaranteed trajectory planning and control based on GP estimation for unmanned surface vessels (Poster) | |
| Learning Contraction Policies from Offline Data (Poster) | |
| ProBF: Probabilistic Safety Certificates with Barrier Functions (Poster) | |
| Parametric-Control Barrier Function-based Adaptive Safe Merging Control for Heterogeneous Vehicles (Poster) | |
| Learning Behavioral Soft Constraints from Demonstrations (Poster) | |
| Execute Order 66: Targeted Data Poisoning for Reinforcement Learning via Minuscule Perturbations (Poster) | |
| Learning Robustly Safe Output Feedback Controllers from Noisy Data with Performance Guarantees (Poster) | |
| State Augmented Constrained Reinforcement Learning: Overcoming the Limitations of Learning with Rewards (Poster) | |
| MESA: Offline Meta-RL for Safe Adaptation and Fault Tolerance (Poster) | |
| Risk Sensitive Model-Based Reinforcement Learning using Uncertainty Guided Planning (Poster) | |
| Adversarial Training Blocks Generalization in Neural Policies (Poster) | |
| Robust Physical Parameter Identification through Global Linearisation of System Dynamics (Poster) | |
| Unbiased Efficient Feature Counts for Inverse RL (Poster) | |
| Bayesian Inverse Constrained Reinforcement Learning (Poster) | |
| Behavior Policy Search for Risk Estimators in Reinforcement Learning (Poster) | |
| Safe Online Exploration with Nonlinear Constraints (Poster) | |
| Avoiding Negative Side Effects by Considering Others (Poster) | |
| Uncertainty-based Safety-Critical Control using Bayesian Methods (Poster) | |
| Reinforcement Learning with Feedback from Multiple Humans with Diverse Skills (Poster) | |
| Specification-Guided Learning of Nash Equilibria with High Social Welfare (Poster) | |
| Conservative and Adaptive Penalty for Model-Based Safe Reinforcement Learning (Poster) | |
| Safe Reinforcement Learning for Grid Voltage Control (Poster) | |