English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Learning a risk-aware policy is essential but rather challenging inunstructured robotic tasks. Safe reinforcement learning methods open up newpossibilities to tackle this problem. However, the conservative policy updatesmake it intractable to achieve sufficient exploration and desirable performancein complex, sample-expensive environments. In this paper, we propose adual-agent safe reinforcement learning strategy consisting of a baseline and asafe agent. Such a decoupled framework enables high flexibility, dataefficiency and risk-awareness for RL-based control. Concretely, the baselineagent is responsible for maximizing rewards under standard RL settings. Thus,it is compatible with off-the-shelf training techniques of unconstrainedoptimization, exploration and exploitation. On the other hand, the safe agentmimics the baseline agent for policy improvement and learns to fulfill safetyconstraints via off-policy RL tuning. In contrast to training from scratch,safe policy correction requires significantly fewer interactions to obtain anear-optimal policy. The dual policies can be optimized synchronously via ashared replay buffer, or leveraging the pre-trained model or thenon-learning-based controller as a fixed baseline agent. Experimental resultsshow that our approach can learn feasible skills without prior knowledge aswell as deriving risk-averse counterparts from pre-trained unsafe policies. Theproposed method outperforms the state-of-the-art safe RL algorithms ondifficult robot locomotion and manipulation tasks with respect to both safetyconstraint satisfaction and sample efficiency.

Safety Correction from Baseline: Towards the Risk-aware Policy in Robotics via Dual-agent Reinforcement Learning