EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning
arXiv cs.CL · 28 Oct 2026, 03:38 AEST
EnvSched is an online scheduler designed to enhance the efficiency of agent reinforcement learning by organizing and evaluating task instances. It achieves this by predicting reward variance and allocating tasks to maximize informative and diverse batches.
Key facts
- As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive.
- To address this question, we introduce EnvSched (Environment-aware Scheduling), an online scheduler that learns from rollout history.
- EnvSched organizes environment--task pairs as nodes in a dual-relation graph, with within-environment edges capturing task semantics and shared tool, state, and verifier structure.
- In RL on Qwen3-8B, EnvSched achieves an average score of 48.81 across BFCL-v3, tau-Bench, and ACEBench-Agent after 40 training steps, exceeding the highest score attained by random task selection after 120 steps.
Read aloud
EnvSched is an online scheduler designed to enhance the efficiency of agent reinforcement learning by organizing and evaluating task instances. It achieves this by predicting reward variance and allocating tasks to maximize informative and diverse batches. As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive. To address this question, we introduce EnvSched Environment-aware Scheduling, an online scheduler that learns from rollout history. EnvSched organizes environment--task pairs as nodes in a dual-relation graph, with within-environment edges capturing task semantics and shared tool, state, and verifier structure. In RL on Qwen3-8B, EnvSched achieves an average score of forty eight point eight one across BFCL-v3, tau-Bench, and ACEBench-Agent after forty training steps, exceeding the highest score attained by random task selection after one hundred twenty steps.
Sources in this cluster
- EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning
arXiv cs.CL · 28 Oct 2026, 03:38 AEST