Tag
This paper introduces DROPJ, a human-centred method for safely training and deploying agent policies by learning a world model from real-world trajectories, then eliciting human preferences with justifications to train a reward model for model predictive control. Experiments show that using human-generated simulated trajectories and justifications improves safety and reduces computational cost.
This paper proposes an uncertainty-aware reinforcement learning framework for autonomous driving that uses expert advice guided by adaptive uncertainty thresholds and a commitment-cooldown strategy to improve safety and efficiency. Experiments in the CARLA simulator show a 5-7% success improvement over the IQN baseline.
OpenAI introduces Safety Gym, a new benchmark environment and toolkit for studying constrained reinforcement learning and safe exploration. The platform features multiple robots and tasks designed to quantify and measure safe exploration through cost functions alongside reward functions.
OpenAI proposes standardizing constrained RL as the formalism for safe exploration and introduces Safety Gym, a benchmark suite for evaluating safe deep RL algorithms in high-dimensional continuous control tasks with safety constraints.