arXiv subjects

Zheng Shen

Publications and source records attributed to Zheng Shen.

4 recordsLinked to original sources

arXiv2026

A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation

Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that since the proximal policy only serves as a trust region anchor between the behavior and target policies, we can approximate it through simple interpolation without explicit computation. We call this approach A-3PO (APproximated Proximal Policy Optimization). A-3PO eliminates this overhead, accelerating training by 1.8x speedup while maintaining comparable performance. Code \& off-the-shelf example are contributed to the open-source RL training system AReaL at: https://github.com/areal-project/AReaL/blob/v1.0.0.rc1/docs/algorithms/prox_approx.md

cs.LG↗

arXiv2026

Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor

MXFP4 arithmetic can dramatically accelerate reinforcement learning (RL) post-training of large language models (LLMs), yet the quantization error introduces severe accuracy degradation. Existing work treats the quantization error as a monolithic noise term, missing the distinct mechanisms upon interpreting how quantization error damages training. We prove an exact three-way decomposition of quantization error and show how each component dominates a distinct RL training pathway. Our theoretical and empirical analysis decomposes the MXFP4 quantization error into three additive components: "scale bias" from power-of-two rounding, "deadzone truncation" from zeroing small values, and "grid noise" from rounding to the nearest 4-bit grid. Each component dominates a distinct RL failure mode: scale bias accumulates multiplicatively through the backward pass, affecting gradient accuracy; deadzone truncation degrades rollout quality; and grid noise raises the policy's entropy. We combine corrections that are RL failure mode-targeted but not component-exclusive: Macro-block scaling to reduce scale bias, Outlier Fallback recovers deadzone entries, but also partially reduces scale bias induced error, and Adaptive Quantization Noise (AQN) for controlling the policy entropy. On Qwen2.5-3B dense and Qwen3-30B-A3B-Base mixture-of-experts model, the targeted corrections recover BF16 accuracy to within 0.7% and exceed BF16 by +1.0% respectively.

cs.LG↗

arXiv2025

LEMMo-Plan: LLM-Enhanced Learning from Multi-Modal Demonstration for Planning Sequential Contact-Rich Manipulation Tasks

Large Language Models (LLMs) have gained popularity in task planning for long-horizon manipulation tasks. To enhance the validity of LLM-generated plans, visual demonstrations and online videos have been widely employed to guide the planning process. However, for manipulation tasks involving subtle movements but rich contact interactions, visual perception alone may be insufficient for the LLM to fully interpret the demonstration. Additionally, visual data provides limited information on force-related parameters and conditions, which are crucial for effective execution on real robots. In this paper, we introduce an in-context learning framework that incorporates tactile and force-torque information from human demonstrations to enhance LLMs' ability to generate plans for new task scenarios. We propose a bootstrapped reasoning pipeline that sequentially integrates each modality into a comprehensive task plan. This task plan is then used as a reference for planning in new task configurations. Real-world experiments on two different sequential manipulation tasks demonstrate the effectiveness of our framework in improving LLMs' understanding of multi-modal demonstrations and enhancing the overall planning performance.

cs.RO↗

arXiv2024

Safe Execution of Learned Orientation Skills with Conic Control Barrier Functions

In the field of Learning from Demonstration (LfD), Dynamical Systems (DSs) have gained significant attention due to their ability to generate real-time motions and reach predefined targets. However, the conventional convergence-centric behavior exhibited by DSs may fall short in safety-critical tasks, specifically, those requiring precise replication of demonstrated trajectories or strict adherence to constrained regions even in the presence of perturbations or human intervention. Moreover, existing DS research often assumes demonstrations solely in Euclidean space, overlooking the crucial aspect of orientation in various applications. To alleviate these shortcomings, we present an innovative approach geared toward ensuring the safe execution of learned orientation skills within constrained regions surrounding a reference trajectory. This involves learning a stable DS on SO(3), extracting time-varying conic constraints from the variability observed in expert demonstrations, and bounding the evolution of the DS with Conic Control Barrier Function (CCBF) to fulfill the constraints. We validated our approach through extensive evaluation in simulation and showcased its effectiveness for a cutting skill in the context of assisted teleoperation.

cs.RO↗