Search papers, labs, and topics across Lattice.
This paper addresses the limitations of contrastive reinforcement learning (CRL) in failure-terminated Markov decision processes by introducing Safe Contrastive Reinforcement Learning (Safe-CRL), which incorporates corrections for the overestimation bias caused by neglecting the probability mass lost to failures. The authors propose two key adjustments: mass-weighted InfoNCE for critic learning and a log-survival-mass score for policy optimization, enabling effective goal-conditioned policy learning using only one-bit failure signals. Experimental results across twelve challenging robot navigation and locomotion tasks demonstrate that Safe-CRL significantly enhances survival rates and outperforms existing CRL methods in achieving goal-reaching performance.
Safe-CRL reveals that even minimal failure signals can effectively guide safe policy learning, transforming how we approach reinforcement learning in high-risk environments.
Contrastive reinforcement learning (CRL) scales effectively in goal-conditioned tasks by casting policy learning into a self-supervised contrastive objective. However, in a failure-terminated Markov decision process, established CRL considers pre-failure future goals only when constructing positive samples, without accounting for the probability mass removed by failure termination. Our theoretical analysis shows that this omission induces a systematic overestimation bias in goal-reaching values. Consequently, near-failure trajectories provide disproportionately strong supervision of success despite retaining little future occupancy. Unsafe actions can thereby be reinforced through catastrophic failure bootstrapping, leading to failed policy learning and unsustainable goal-reaching behaviours. To address this problem, we introduce two minimal yet strong corrections: mass-weighted InfoNCE corrects the overweighting of short surviving futures in critic learning, and a log-survival-mass score restores the missing survival mass in policy optimization. The resulting method, Safe Contrastive Reinforcement Learning (Safe-CRL), requires only the one-bit signal provided by failure termination to scale safe goal-conditioned policy learning. Across twelve failure-prone robot navigation and locomotion tasks, Safe-CRL consistently improves survival and substantially outperforms the Scaling-CRL baseline in goal-reaching performance. Additionally, deep Safe-CRL policies exhibit complex failure-avoidance behaviours. This study completes the CRL theory under failure termination and provides a scalable safe RL framework. The code is available via https://github.com/RomainLITUD/safe-crl.
