arXiv subjects

Jinming Ma

Publications and source records attributed to Jinming Ma.

15 recordsLinked to original sources

arXiv2026

HiFi-UMI: Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone

Learning deployable manipulation policies is bottlenecked by the scarcity of data that is both high-fidelity and scalable. Real-robot teleoperation is accurate but costly to scale; robot-free UMI capture scales readily, and current practice uses the resulting data mainly for pre-training, adding a small real-robot "anchor" at post-training. We ask whether raising the fidelity of robot-free UMI data, rather than shrinking the real-robot fraction, can remove that anchor. We present HiFi-UMI, a portable UMI data-production system co-designed for trajectory accuracy, inter-gripper relative pose, synchronization, and field of view: head-mounted offline stereo-inertial SLAM, native rather than reconstructed relative pose, a shared microsecond GPIO trigger, and two wide-angle cameras per hand covering ~200 degrees. It reaches 3 mm workspace-local end-effector accuracy without external tracking infrastructure. Using this corpus, we demonstrate zero-robot post-training: a policy post-trained solely on HiFi-UMI demonstrations deploys directly on a real robot and matches in-domain teleoperation across three backbones spanning the vision-language-action and world-action-model families, with success-rate differences of -2.5, +3.1, and -0.6 percentage points on StarVLA-QwenPI, OpenPI-pi_0.5, and LingBot-VA; the strongest policy reaches 85% on a precision insertion task, even though the teleoperation baseline is collected in the evaluation scene and no HiFi-UMI trajectory is. Pre-training on 4,000 hours from the same corpus lowers action error on ten unseen tasks by 41% and, on StarVLA-QwenPI, raises real-robot success by a further 18.1 percentage points. We open-source HiFi-UMI-2K, 2,000 hours of microsecond-synchronized, ultra-wide-FoV demonstrations, each automatically reconstructed and validated through simulation replay, as a large-scale, high-fidelity resource for the robot-learning community.

cs.RO↗

arXiv2026

Diagnosing Semantic Handoff Failures in Agent-Orchestrated Vision-Language-Action Skill Composition

Long-horizon household tasks require robots to compose many language-conditioned skills, yet the boundary between consecutive skills is rarely explicit. A skill may satisfy its own postcondition while leaving the robot, objects, or camera views in a state from which the next skill cannot reliably start. We study this semantic handoff problem in BEHAVIOR-1K through an agent-orchestrated vision-language-action execution harness. The harness invokes $\pi_{0.5}$-based skill checkpoints trained from cleaned BEHAVIOR-1K demonstrations, assigns each skill typed arguments and a step budget, and uses multi-view vision-language model verification to decide whether execution should advance, retry, or replan. To separate isolated skill competence from long-horizon compositional robustness, we evaluate the same checkpoints under two initial-state distributions: clean skill-boundary snapshots and chained terminal states produced by previous skills. Selected navigation, grasping, placement, and door-opening skills achieve 77--100% success from clean snapshots under human-reviewed verification, yet composed rollouts still frequently stall from chained states. The resulting traces attribute failures to next-skill readiness, target grounding, and control execution, turning nearzero task success into actionable diagnostics for what VLA skill libraries must learn next: robustness to the messy chained-state distribution that clean demonstrations underrepresent.

cs.RO↗

arXiv2026

I\textsuperscript{2}RiMA: Spectral Riemannian Representation with Temporal Attention for Mental Stress Detection based on EEG Signals

Cross-subject EEG stress detection remains challenging because discriminative stress-related patterns are both subject-dependent and frequency-specific. Conventional Riemannian methods model spatial covariance mainly in the time domain, overlooking neural oscillations that are critical for high-level cognitive state decoding, while standard temporal tokenization often fragments inter-slice temporal coherence. To address these limitations, we propose \method{}, an Intra-Inter Riemannian Manifold Attention Network for EEG-based stress detection. \method{} constructs spatial covariance matrices independently at each frequency point and maps them to the SPD tangent space, preserving channel-wise geometry together with frequency-specific discriminative cues. It further introduces frequency cluster aggregation to select informative spectral components and reduce redundancy by forming compact, data-driven frequency clusters aligned with EEG rhythms. Finally, an intra-inter slice attention module adaptively integrates local slice-level spectral dynamics and global temporal context across EEG sequences. Experiments on three datasets show that \method{} consistently outperforms five state-of-the-art baselines, achieving up to 82.78\% balanced accuracy while remaining efficient with only 1.60M parameters and 31.95M FLOPs.

cs.LG↗

arXiv2026

Tropical: Enhancing SLO Attainment in Disaggregated LLM Serving via SLO-Aware Multiplexing

To guarantee service quality in transformer based large language model (LLM) serving, it is essential to meet the latency constraints of both the prefill phase (measured by Time-to-First-Token, TTFT) and the decode phase (measured by Time-per-Output-Token, TPOT). Non-disaggregated serving places prefill and decode on the same worker, while disaggregated serving places the prefill and decode on isolated workers. However, no single architecture excels in both TTFT and TPOT metrics. After conducting a root cause analysis, we concluded that indisaggregated LLM serving, prefill execution has minimal interference with decode execution but result in high queuing times. In contrast,non-disaggregated LLM serving effectively reduces queuing times but introduces significant interference between prefills and decodes. In order to leverage the best aspects of both non-disaggregated anddisaggregated LLM serving, we have designed and implemented Tropical.Tropical introduces an sevice-level objectives (SLO)-aware multiplexing strategy that balances the queuing time and the interference, enabling the LLM serving to achieve high TTFT and TPOT SLOs simultaneously. Our evaluation of real-world datasets reveals that Tropical outperforms both state-of-the-art non-disaggregated and disaggregated LLM serving systems, achieving up to 2.09 more requests within a 90% SLO attainment. Specially, compared to the disaggregated LLM serving system, Tropicalimproves P90 TTFT performance by 9 with only an 15% reduction in P90 TPOT. Against the non-disaggregated LLM serving systems, Tropicaldelivers a 2.8 performance improvement in P90 TPOT while maintaining the same P90 TTFT.

cs.DC↗

arXiv2026

NanoCP: Request-Level Dynamic Context Parallelism for Data-Expert Parallel Decoding

Modern serving systems for Mixture-of-Experts (MoE) models adopt hybrid data-expert parallelism: expert parallelism (EP) shards experts across GPUs to scale capacity, while data parallelism (DP) replicates attention layers across instances to process independent requests. Existing systems bind each request's attention, MoE communication, and KV cache to a single instance. Because attention latency scales with KV cache size while MoE communication latency scales with batch size, this binding cannot balance both simultaneously, producing EP stragglers; it also fragments KV memory across instances, inflating tail latency under long contexts. While existing context parallelism (CP) mitigates these constraints, its uniform parallelism degree incurs prohibitive communication and attention-side overheads. We present \work, which decouples MoE communication from KV cache placement and achieves dual balance through dynamic context parallelism (DCP). DCP assigns each request a context-parallel degree sized to its KV footprint: long requests distribute attention across multiple instances; short requests remain local. This dynamic parallelism effectively liquefies the KV cache across the cluster, balancing both the per-instance KV cache occupancy and batch sizes without unnecessary load-balancing costs. To bridge DCP with static execution, \work introduces an ahead-of-time (AOT) graph engine paired with a custom routing-based communication backend. Experimental results show that \work maintains up to $1.88\times$--$3.27\times$ higher request rates under strict time-per-output-token (TPOT) service level objectives (SLOs). Furthermore, \work significantly mitigates stragglers, reducing P99 tail latency by up to $1.79\times$--$2.12\times$.

cs.DC↗

arXiv2026

RL-Based Coverage Path Planning for Deformable Objects on 3D Surfaces

Currently, manipulation tasks for deformable objects often focus on activities like folding clothes, handling ropes, and manipulating bags. However, research on contact-rich tasks involving deformable objects remains relatively underdeveloped. When humans use cloth or sponges to wipe surfaces, they rely on both vision and tactile feedback. Yet, current algorithms still face challenges with issues like occlusion, while research on tactile perception for manipulation is still evolving. Tasks such as covering surfaces with deformable objects demand not only perception but also precise robotic manipulation. To address this, we propose a method that leverages efficient and accessible simulators for task execution. Specifically, we train a reinforcement learning agent in a simulator to manipulate deformable objects for surface wiping tasks. We simplify the state representation of object surfaces using harmonic UV mapping, process contact feedback from the simulator on 2D feature maps, and use scaled grouped convolutions (SGCNN) to extract features efficiently. The agent then outputs actions in a reduced-dimensional action space to generate coverage paths. Experiments demonstrate that our method outperforms previous approaches in key metrics, including total path length and coverage area. We deploy these paths on a Kinova Gen3 manipulator to perform wiping experiments on the back of a torso model, validating the feasibility of our approach.

cs.RO↗

arXiv2026

Choose What to Manipulate: Revealing Data Scaling Laws in Bounding-Box Guided Policies for Semantic Manipulation

Diffusion-based policies generalize poorly in semantic manipulation, a key obstacle to real-world deployment, because text-only instructions cannot reliably steer the policy toward the target object in cluttered, dynamic scenes. We instead use bounding-box instructions to specify the target directly, and study how performance scales with data. To this end, we build Label-UMI, a handheld segmentation device with an automated annotation pipeline for efficiently collecting semantically labeled demonstrations, and propose a semantic-motion-decoupled framework that couples object detection with a bounding-box-guided diffusion policy; a first-frame anchoring mechanism keeps execution robust to missed detections and noisy boxes. We find that generalization follows a bounded, saturating data-scaling law with diminishing returns, validated on four real-world tasks with 6,400 demonstrations, and distill an object-diversity-first collection strategy reaching 85\% success in cluttered scenes. All data and code will be released.

cs.RO↗

arXiv2026

An Efficient and Multi-Modal Navigation System with One-Step World Model

Navigation is a fundamental capability for mobile robots. While the current trend is to use learning-based approaches to replace traditional geometry-based methods, existing end-to-end learning-based policies often struggle with 3D spatial reasoning and lack a comprehensive understanding of physical world dynamics. Integrating world models-which predict future observations conditioned on given actions-with iterative optimization planning offers a promising solution due to their capacity for imagination and flexibility. However, current navigation world models, typically built on pure transformer architectures, often rely on multi-step diffusion processes and autoregressive frame-by-frame generation. These mechanisms result in prohibitive computational latency, rendering real-time deployment impossible. To address this bottleneck, we propose a lightweight navigation world model that adopts a one-step generation paradigm and a 3D U-Net backbone equipped with efficient spatial-temporal attention. This design drastically reduces inference latency, enabling high-frequency control while achieving superior predictive performance. We also integrate this model into an optimization-based planning framework utilizing anchor-based initialization to handle multi-modal goal navigation tasks. Extensive closed-loop experiments in both simulation and real-world environments demonstrate our system's superior efficiency and robustness compared to state-of-the-art baselines.

cs.RO↗

arXiv2026

Learning Diverse Skills for Behavior Models with Mixture of Experts

Imitation learning has demonstrated strong performance in robotic manipulation by learning from large-scale human demonstrations. While existing models excel at single-task learning, it is observed in practical applications that their performance degrades in the multi-task setting, where interference across tasks leads to an averaging effect. To address this issue, we propose to learn diverse skills for behavior models with Mixture of Experts, referred to as Di-BM. Di-BM associates each expert with a distinct observation distribution, enabling experts to specialize in sub-regions of the observation space. Specifically, we employ energy-based models to represent expert-specific observation distributions and jointly train them alongside the corresponding action models. Our approach is plug-and-play and can be seamlessly integrated into standard imitation learning methods. Extensive experiments on multiple real-world robotic manipulation tasks demonstrate that Di-BM significantly outperforms state-of-the-art baselines. Moreover, fine-tuning the pretrained Di-BM on novel tasks exhibits superior data efficiency and the reusable of expert-learned knowledge. Code is available at https://github.com/robotnav-bot/Di-BM.

cs.RO↗

arXiv2025

Reinforced Embodied Planning with Verifiable Reward for Real-World Robotic Manipulation

Enabling robots to execute long-horizon manipulation tasks from free-form language instructions remains a fundamental challenge in embodied AI. While vision-language models (VLMs) have shown promise as high-level planners, their deployment in the real world is hindered by two gaps: (i) the scarcity of large-scale, sequential manipulation data that couples natural language with multi-step action plans, and (ii) the absence of dense, interpretable rewards for fine-tuning VLMs on planning objectives. To address these issues, we propose REVER, a framework that empowers VLMs to generate and validate long-horizon manipulation plans from natural language instructions in real-world scenarios. Under REVER we train and release RoboFarseer, a VLM incentivized to emit chain-of-thought that perform temporal and spatial reasoning, ensuring physically plausible and logically coherent plans. To obtain training data, we leverage the Universal Manipulation Interface framework to capture hardware-agnostic demonstrations of atomic skills. An automated annotation engine converts each demonstration into vision-instruction-plan triplet. We introduce a verifiable reward that scores the generated plan by its ordered bipartite matching overlap with the ground-truth skill sequence. At run time, the fine-tuned VLM functions both as a planner and as a monitor, verifying step-wise completion. RoboFarseer matches or exceeds the performance of proprietary models that are orders of magnitude larger, while on open-ended planning it surpasses the best baseline by more than 40%. In real-world, long-horizon tasks, the complete system boosts overall success by roughly 60% compared with the same low-level controller without the planner. We will open-source both the dataset and the trained model upon publication.

cs.RO↗

arXiv2024

Safety Constrained Multi-Agent Reinforcement Learning for Active Voltage Control

Active voltage control presents a promising avenue for relieving power congestion and enhancing voltage quality, taking advantage of the distributed controllable generators in the power network, such as roof-top photovoltaics. While Multi-Agent Reinforcement Learning (MARL) has emerged as a compelling approach to address this challenge, existing MARL approaches tend to overlook the constrained optimization nature of this problem, failing in guaranteeing safety constraints. In this paper, we formalize the active voltage control problem as a constrained Markov game and propose a safety-constrained MARL algorithm. We expand the primal-dual optimization RL method to multi-agent settings, and augment it with a novel approach of double safety estimation to learn the policy and to update the Lagrange-multiplier. In addition, we proposed different cost functions and investigated their influences on the behavior of our constrained MARL method. We evaluate our approach in the power distribution network simulation environment with real-world scale scenarios. Experimental results demonstrate the effectiveness of the proposed method compared with the state-of-the-art MARL methods. This paper is published at \url{https://www.ijcai.org/Proceedings/2024/}.

cs.LG↗

arXiv2024

Improving Offline Reinforcement Learning with Inaccurate Simulators

Offline reinforcement learning (RL) provides a promising approach to avoid costly online interaction with the real environment. However, the performance of offline RL highly depends on the quality of the datasets, which may cause extrapolation error in the learning process. In many robotic applications, an inaccurate simulator is often available. However, the data directly collected from the inaccurate simulator cannot be directly used in offline RL due to the well-known exploration-exploitation dilemma and the dynamic gap between inaccurate simulation and the real environment. To address these issues, we propose a novel approach to combine the offline dataset and the inaccurate simulation data in a better manner. Specifically, we pre-train a generative adversarial network (GAN) model to fit the state distribution of the offline dataset. Given this, we collect data from the inaccurate simulator starting from the distribution provided by the generator and reweight the simulated data using the discriminator. Our experimental results in the D4RL benchmark and a real-world manipulation task confirm that our method can benefit more from both inaccurate simulator and limited offline datasets to achieve better performance than the state-of-the-art methods.

cs.RO↗

arXiv2023

Real-Time Network-Level Traffic Signal Control: An Explicit Multiagent Coordination Method

Efficient traffic signal control (TSC) has been one of the most useful ways for reducing urban road congestion. Key to the challenge of TSC includes 1) the essential of real-time signal decision, 2) the complexity in traffic dynamics, and 3) the network-level coordination. Recent efforts that applied reinforcement learning (RL) methods can query policies by mapping the traffic state to the signal decision in real-time, however, is inadequate for unexpected traffic flows. By observing real traffic information, online planning methods can compute the signal decisions in a responsive manner. We propose an explicit multiagent coordination (EMC)-based online planning methods that can satisfy adaptive, real-time and network-level TSC. By multiagent, we model each intersection as an autonomous agent, and the coordination efficiency is modeled by a cost (i.e., congestion index) function between neighbor intersections. By network-level coordination, each agent exchanges messages with respect to cost function with its neighbors in a fully decentralized manner. By real-time, the message passing procedure can interrupt at any time when the real time limit is reached and agents select the optimal signal decisions according to the current message. Moreover, we prove our EMC method can guarantee network stability by borrowing ideas from transportation domain. Finally, we test our EMC method in both synthetic and real road network datasets. Experimental results are encouraging: compared to RL and conventional transportation baselines, our EMC method performs reasonably well in terms of adapting to real-time traffic dynamics, minimizing vehicle travel time and scalability to city-scale road networks.

cs.AI↗

arXiv2023

Effective Multimodal Reinforcement Learning with Modality Alignment and Importance Enhancement

Many real-world applications require an agent to make robust and deliberate decisions with multimodal information (e.g., robots with multi-sensory inputs). However, it is very challenging to train the agent via reinforcement learning (RL) due to the heterogeneity and dynamic importance of different modalities. Specifically, we observe that these issues make conventional RL methods difficult to learn a useful state representation in the end-to-end training with multimodal information. To address this, we propose a novel multimodal RL approach that can do multimodal alignment and importance enhancement according to their similarity and importance in terms of RL tasks respectively. By doing so, we are able to learn an effective state representation and consequentially improve the RL training process. We test our approach on several multimodal RL domains, showing that it outperforms state-of-the-art methods in terms of learning speed and policy quality.

cs.LG↗

arXiv2022

Feudal Multi-Agent Reinforcement Learning with Adaptive Network Partition for Traffic Signal Control

Multi-agent reinforcement learning (MARL) has been applied and shown great potential in multi-intersections traffic signal control, where multiple agents, one for each intersection, must cooperate together to optimize traffic flow. To encourage global cooperation, previous work partitions the traffic network into several regions and learns policies for agents in a feudal structure. However, static network partition fails to adapt to dynamic traffic flow, which will changes frequently over time. To address this, we propose a novel feudal MARL approach with adaptive network partition. Specifically, we first partition the network into several regions according to the traffic flow. To do this, we propose two approaches: one is directly to use graph neural network (GNN) to generate the network partition, and the other is to use Monte-Carlo tree search (MCTS) to find the best partition with criteria computed by GNN. Then, we design a variant of Qmix using GNN to handle various dimensions of input, given by the dynamic network partition. Finally, we use a feudal hierarchy to manage agents in each partition and promote global cooperation. By doing so, agents are able to adapt to the traffic flow as required in practice. We empirically evaluate our method both in a synthetic traffic grid and real-world traffic networks of three cities, widely used in the literature. Our experimental results confirm that our method can achieve better performance, in terms of average travel time and queue length, than several leading methods for traffic signal control.

cs.MA↗