Safe RL
PulseAugur coverage of Safe RL — every cluster mentioning Safe RL across labs, papers, and developer communities, ranked by signal.
-
New attack exploits demonstrations to compromise safe RL robotic controllers
Researchers have developed a new method to attack black-box safe reinforcement learning (Safe RL) controllers used in robotics. This demonstration-guided observation attack can identify safety violations by recovering a…
-
New benchmark reveals reset-free RL agents struggle with irreversible actions
Researchers have introduced REVERSAL-BENCH, a new benchmark designed to measure the limitations of reset-free reinforcement learning agents in environments with irreversible actions. The benchmark controls the degree of…
-
New algorithm enables safe transfer of AI policies from simulation to real-world
Researchers have developed a novel algorithm for safe sim-to-real transfer in reinforcement learning, addressing the challenge of deploying policies trained in simulators to the real world. The algorithm leverages simul…
-
New method enhances safety in hierarchical reinforcement learning tasks
Researchers have developed a novel method to enhance safety in hierarchical reinforcement learning, particularly for complex, long-horizon tasks. The approach utilizes a learned world model combined with a high-level po…
-
New benchmark tests AI's hierarchical moral alignment in ethical dilemmas
Researchers have developed MoralityGym, a new benchmark designed to evaluate how well AI agents can navigate complex ethical dilemmas and adhere to hierarchical moral norms. The benchmark utilizes a novel formalism call…