Tag
This paper studies the sample complexity of policy learning under the mu-resets interaction protocol in reinforcement learning, resolving a question about the role of policy realizability and showing horizon dependence is exponential under all-policy concentrability and sqrt-exponential under pushforward concentrability.
Introduces EvoHarness-RL, a framework that learns runtime harness policies for long-horizon LLM agents, enabling them to construct and update external state (belief, progress, experience) during task execution. Using Qwen3-8B on ALFWorld, it achieves 96.9% success and reveals harness annealing and evolution dynamics.
This paper proposes StructPO, a structure-aware policy learning framework that internalizes multi-stage academic writing workflows into a single-pass LLM policy using explicit stage tokens and refinement-guided optimization, improving introduction generation quality and efficiency.
This paper proposes SG-WAM, a self-guided framework for learning geometry-aware action-conditioned world models directly in policy-derived representation space. It achieves state-of-the-art success rates on LIBERO and LIBERO-Plus benchmarks, outperforming strong baselines in real-world evaluations.
HiFi-UMI introduces a portable data-production system for robot-free UMI data that achieves high trajectory accuracy using stereo-inertial SLAM and wide-angle cameras. Training manipulation policies on this data alone enables zero-shot deployment on real robots, matching or exceeding teleoperation baselines across several model families, and the authors open-source a 2,000-hour high-fidelity dataset.
This paper investigates why text-based optimization (TextGrad) fails for language agents, showing that while frozen agents can follow good policies, they cannot reliably learn and select policies from their own trajectories.
Proposes a graph-constrained traversal policy that reformulates ICD-10-CM code prediction as a finite-horizon decision process over a pruned code hierarchy, outperforming flat baselines on MIMIC-IV discharge summaries.
BlockPilot proposes an instance-adaptive policy that predicts the optimal block size for diffusion-based speculative decoding, achieving significant speedup with minimal overhead.
SimFoundry is a modular system that automates real-to-sim scene construction from video, generating digital twins and affordance-preserving variations for zero-shot robot policy training, achieving strong transfer to real-world tasks and high simulation-to-real performance prediction.
The paper presents World Value Model (WVM), a generalist robotic value model that combines world models with value estimation to accurately assess task progression and improve robotic policy learning from mixed-quality data, achieving state-of-the-art results on standard benchmarks and a new suboptimal data benchmark.
This paper from MIT and Harvard explores automating SKILL.md generation by mining GUI interaction trajectories, finding that clusters are readable but do not improve policy performance across domains.
PoLAR introduces a geometrically structured latent action representation in hyperbolic space that separates transition extent from mode, improving robotic policy learning performance.
Geometric Action Model repurposes a geometric foundation model for robot policy learning, achieving 85.5% on LIBERO-Plus with 6.9 ms inference, 55× faster than baselines.
The Geometric Action Model (GAM) repurposes a pretrained geometric foundation model (GFM) as a unified backbone for language-conditioned robot manipulation, achieving higher accuracy, robustness, and efficiency than existing foundation-model-scale baselines across simulation and real-world benchmarks.
Light-WAM is a lightweight world action model for efficient robot manipulation that uses a compact video backbone and downsampled latent space for future-video supervision, achieving high performance with low inference latency.
DiffAero is a GPU-accelerated, fully differentiable simulation framework for quadrotor control policy learning that supports environment- and agent-level parallelism, multiple dynamics models, and customizable sensors. It enables robust flight policy learning in hours on consumer-grade hardware and is released as open-source.
This paper introduces Capability Self-Assessment (CSA) for LLMs, formulating it as a policy-learning problem. Experiments show that reinforcement learning effectively teaches models to recognize their own limits and delegate queries they cannot solve, outperforming supervised fine-tuning and generalizing well out-of-distribution.
τ_0-WM is a unified video-action world model for robotic manipulation that integrates policy learning, video prediction, and action evaluation using a shared video diffusion backbone. It shows superior performance on challenging long-horizon and fine-grained tasks.
WorldString is a neural architecture that models object state manifolds from point clouds or RGB-D video streams, serving as a foundational component for physical world models with differentiable structure for policy learning integration.
This paper introduces HCL-GP, a dynamic policy-learning framework that integrates generalized planning and hierarchical task decomposition to enable LLM-based agents to learn and reuse executable policy components, significantly improving performance on the AppWorld benchmark.