Allocates token-level credit in rubric-conditioned GRPO by counterfactually replaying the same response under rubric and criteria-free prompts, using tokenwise log-likelihood contrasts to compute bounded, response-normalized weights that redistribute GRPO advantages without training an auxiliary scorer.
Regularizes latent world models by replacing the Epps–Pulley Gaussianization objective with a quantile–quantile matching loss that aligns projected latent samples to rank-matched Gaussian quantiles, improving tail correction and planning success via cross-batch ranking.
AI Agent Papers2026
Evaluates vision-language model judges on computer-using agent (CUA) trajectories to measure verifier reliability. Provides OSReward-Hard and OSReward-Multi challenge sets, the OS-Shepherd-100K reasoning-annotated corpus, and trained OS-Shepherd reward models that match commercial judges at ~30–60× lower cost.
Frames skill generation as a sequential editing task and introduces a novel rollback reward to train an RL generator (Skill-α) that evaluates each edit by its downstream execution impact, producing skills that improve agent success rates across document-to-skill and experience-to-skill settings.
Supervises audio reasoning by generating per-sample, audio-grounded rubrics that evolve with model rollouts and serve as reinforcement-learning rewards, improving perception and adaptive multi-step reasoning while avoiding reward saturation.
Analyzes why supervised fine-tuning (SFT) causes severe task conflicts under multi-stage multi-task training while reinforcement learning (RL) enables stable coexistence, attributing the effect to sparse, near-orthogonal RL parameter updates and proposing Parallel-RL to decouple multi-task training.
AI Agent Papers2026
Analyzes how to build effective training environment distributions for multimodal agents and proposes Ability-aware Environment Selection (AES) and Hierarchical Difficulty Curriculum (HDC) to improve diversity and difficulty scheduling, yielding large relative gains in experiments.
Replaces external environment interaction in agentic RL training with 'world rehearsal': the policy alternates between making tool calls and simulating their environment responses, jointly optimizing both roles so the agent internalizes environment dynamics and improves long-horizon tool use and transfer.
AI Agent Papers2026
Converts sparse trajectory-level rewards into turn-level credit by aggregating token-level teacher–student log-probability gaps and recursively updating a Bayesian belief in log-odds; produces turn-wise reweighting for policy optimization without an extra critic or rollouts.
Presents a unified black-box reinforcement learning framework to train and optimize agents running inside complex execution harnesses. Uses sandbox-parallel rollouts, a serving proxy that captures model calls and reconstructs multi-turn trajectories as prefix trees, and adapted GRPO/PPO optimizers to achieve stable, scalable RL across heterogeneous harnesses.
Uses cooperative multi-agent RL where multiple decoupled models provide peer-derived pseudo-rewards to each other, enabling unsupervised improvements in reasoning; increases cohort diversity to reduce correlated errors and avoid training collapse, showing consistent gains across text and multimodal benchmarks.
AI Agent Papers2026
Wraps static, hand-built environments with a programmable plug-in harness that reshapes environment behavior without changing underlying logic. EnvRigger automates diagnosis and synthesis of harness components from agent failure trajectories, validating edits via fresh rollouts to improve agent success and efficiency.