Research→Official→arXiv Robotics
A new system-level acceleration strategy for Vision-Language-Action (VLA) models reduces temporal redundancy in both perception and action generation. The approach incrementally updates only dynamic scene tokens and compresses diffusion sampling to a 2-step schedule, resulting in over 2x speedup while maintaining high performance, with up to 98% success rate on manipulation benchmarks. The method was validated on multiple robotic platforms, including real robots.
Why it matters: This work offers a practical advance for real-time deployment of VLA models in robotics by significantly improving inference speed without sacrificing accuracy.
Jul 15, 2026
Research→Official→arXiv Robotics
Researchers introduce DECO, a decoupled multimodal diffusion transformer designed to disentangle and integrate vision, proprioception, and tactile signals for bimanual dexterous manipulation. Evaluated on the new DECO-50 dataset with real dual-arm robots, DECO achieves a 72.25% average success rate across tasks, outperforming baselines by 21%. A lightweight tactile adapter further improves success rates by 10.25% while requiring less than 10% of parameters to be tuned.
Why it matters: This work demonstrates a significant advance in robotic manipulation by efficiently integrating tactile sensing with other modalities, leading to improved performance on complex bimanual tasks.
Jul 15, 2026
Research→Official→arXiv Robotics
Researchers introduce Jetson-PI, a method for deploying Vision-Language-Action (VLA) models on low-power devices such as the Jetson Orin. Jetson-PI uses foresight-aligned asynchronous inference, including a future correction module and confidence-based scheduling, to address latency and misalignment issues. Experiments show Jetson-PI achieves 8.66x and 5.41x improvements in control frequency over naive PyTorch and vla.cpp, respectively, and outperforms VLASH by 14.8% in success rate on the LIBERO benchmark.
Why it matters: This work makes advanced VLA models practical for real-time robot control on resource-constrained edge hardware.
Jul 15, 2026
Research→Official→arXiv Robotics
DenseReward is a dense robotic reward model designed to overcome the limitations of current vision-language reward models in reinforcement learning for robotics. By introducing an automated pipeline that synthesizes diverse, physically realistic failure trajectories in simulation, DenseReward enables fine-grained, frame-level reward prediction from visual observations and language instructions. Experimental results indicate that DenseReward outperforms existing models in both simulated and real-world robotic manipulation tasks, and provides effective reward signals for reinforcement learning and model predictive control.
Why it matters: This work addresses a major challenge in robotic reinforcement learning by enabling dense, informative reward signals without requiring costly human labeling, potentially accelerating progress in autonomous robot learning.
Jul 15, 2026
Research→Official→arXiv Robotics
Researchers have introduced FEP-Nav, a biologically inspired framework for real-time perceptual adaptation in visual navigation. Drawing on the Free Energy Principle, FEP-Nav reduces prediction error and Bayesian surprise during inference without requiring gradient updates, enabling robust navigation under visual corruptions. Experiments demonstrate that FEP-Nav restores performance lost to visual corruption and outperforms both non-adaptive and strong adaptive baselines in simulated and real-world settings.
Why it matters: This work shows that variational principles can inform the design of robust autonomous navigation systems capable of adapting to degraded sensory input, with implications for robotics and autonomous vehicles.
Jul 15, 2026
Research→Official→arXiv Robotics
FlowWAM introduces a dual-stream diffusion framework that adopts optical flow as a unified, video-native action representation for World Action Models (WAMs). The approach enables leveraging large-scale, action-unlabeled video datasets for pretraining and achieves state-of-the-art results on RoboTwin manipulation (92.94% success rate) and WorldArena world modeling (63.71 EWMScore).
Why it matters: This work bridges video generation and robot control by leveraging optical flow, enabling improved action prediction and world modeling through pretraining on unlabeled video data.
Jul 15, 2026
Research→Official→arXiv Robotics
VistaVLA is a two-stage framework that builds a geometry- and semantics-aware 3D cognitive representation from 3D Gaussian primitives for robotic manipulation. Using a Merge-then-Query mechanism, it compresses dense Gaussian primitives into compact tokens, achieving a 99% reduction in token count while preserving action-relevant 3D layouts and semantic context. In real-world tests, VistaVLA improved success rates by 22.8% across seven tasks and by 30.0% on out-of-distribution tasks compared to the VLA-Adapter baseline.
Why it matters: VistaVLA advances robotic manipulation by enabling explicit 3D semantic reasoning, leading to significantly improved performance in both standard and out-of-distribution real-world scenarios.
Jul 15, 2026
Research→Official→arXiv Robotics
Researchers have developed an automated workflow to optimize and deploy vision-based convolutional neural networks (CNNs) on nano-sized UAVs, specifically the Crazyflie 2.1 drone. Their approach achieves a 2x reduction in memory usage and a 1.6x speedup in inference time for the PULP-Dronet network, enabling improved obstacle avoidance and free flight speeds, all while consuming less than 1.6% of the drone's power budget. The software is open-sourced to encourage further research and applications.
Why it matters: This work demonstrates a significant advance in automating the deployment of efficient neural networks on highly resource-constrained nano-drones, enabling better autonomous navigation and reducing manual development effort.
Jul 15, 2026
Research→Official→arXiv Robotics
Researchers introduce Action Map Policy (AMP), a method that formulates 3D closed-loop manipulation policy learning as a classification problem in image space. By projecting 3D actions onto camera image planes and treating each pixel as a discrete class, AMP enables millimeter-level precision without requiring a large action vocabulary. Experiments demonstrate that AMP achieves higher success rates and faster inference than strong baseline methods across various manipulation tasks.
Why it matters: This approach offers a novel and efficient action representation for robot learning, potentially enabling more precise and scalable manipulation policies.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce Adversarial World Modeling (AWM), a multi-agent self-play fine-tuning framework that transforms a planner's world model into an adversary to generate rare and safety-critical driving scenarios. The approach employs a decoupled min-max game solver with counterfactual credit assignment and regret-aware optimization. Experiments on nuPlan and InterPlan benchmarks indicate that AWM improves closed-loop performance in both typical and challenging long-tail traffic scenarios.
Why it matters: This work provides a principled method for training autonomous vehicles to handle rare and dangerous traffic situations without relying on external scenario generators or extensive simulation.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce PIER-Flow, a lightweight navigation policy for mobile robots that distills a Model Predictive Control (MPC) expert into a continuous-time ODE, enabling single-step action generation. In simulation, PIER-Flow achieves a 98.85% success rate with zero collisions and an average inference time of approximately 1.29 ms, representing a 37.2× speedup over MPC and over 800× over standard diffusion models. Real-world deployment on edge hardware demonstrates stable inference latency of around 5.3 ms, outperforming planning baselines in responsiveness and reliability.
Why it matters: PIER-Flow demonstrates a significant advance in enabling real-time, collision-free navigation for mobile robots in dense environments, particularly on resource-constrained hardware.
Jul 14, 2026
Research→Official→arXiv Robotics
SUREFlow is a new framework for robot manipulation that leverages a Mamba backbone to jointly predict action velocities and input-dependent residual uncertainty, allowing selective refinement of unreliable action dimensions. On the LIBERO benchmark, SUREFlow achieves a 92.5% average success rate, outperforming the Mamba-based MaIL by 34.2%. On LIBERO-PRO, it attains around 49% success rate with only 179M parameters, comparable to much larger vision-language-action models with 3-7B parameters.
Why it matters: SUREFlow shows that explicit uncertainty modeling in action generation can substantially improve the robustness and efficiency of robot manipulation, achieving strong performance with far fewer parameters than existing large models.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce BucketKD, a knowledge distillation framework designed to compress large end-to-end motion planning models for autonomous driving while maintaining safety. The method uses adaptive buckets to capture scene semantics and incorporates a safety-aware attention mechanism based on time-to-collision. Experiments in the CARLA simulator demonstrate that BucketKD achieves higher planning accuracy and safety compared to state-of-the-art methods, with strong model compression.
Why it matters: This approach could facilitate the deployment of safety-critical autonomous driving models on resource-constrained platforms without compromising performance.
Jul 14, 2026
Models→Official→arXiv Robotics
Researchers have introduced SLIDER, a framework for aerial robots that enables efficient target search in large, unknown environments without relying on dense global maps. SLIDER combines a local sliding map with sparse global history, a novel observation quality evaluation, and incremental viewpoint clustering to improve real-time decision-making and reduce computational load. Simulations and real-world experiments show that SLIDER outperforms state-of-the-art methods in memory usage, decision latency, and search efficiency.
Why it matters: This approach could make aerial robots more practical for large-scale search tasks by improving efficiency and reducing hardware requirements.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce Artificial Foveated Perception (AFP), a lightweight, policy-agnostic module that predicts task-conditioned masks over relevant objects and robot parts to address shortcut learning in robotic foundation models. AFP is used as an auxiliary grounding signal during fine-tuning, aligning policy attention with task-relevant regions and improving generalization. The approach does not require AFP at inference time and is shown to reduce fine-tuning time, suppress overfitting, and enhance robustness to environmental perturbations across state-of-the-art models.
Why it matters: This work offers a practical method to mitigate shortcut learning, a key challenge in deploying robust and generalizable robotic foundation models in real-world environments.
Jul 14, 2026
Research→Official→arXiv Robotics
A new algorithm, Inter-POMDP, addresses the challenge of multi-object search in unknown, multi-room household environments by decomposing the problem into two interacting planning levels: a high-level POUCT planner using LLM-informed beliefs about object locations, and a low-level motion planner that accounts for navigation uncertainty with obstacle-aware particle beliefs. Experiments in both simulation and real-world settings demonstrate that Inter-POMDP reduces collisions by up to 63%, navigation steps by up to 35%, and detection counts by up to 32% compared to baseline methods.
Why it matters: This work represents a significant advance in autonomous robot planning, enabling more efficient and safer multi-object search in complex, real-world household environments.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce VINE, a reinforcement learning-oriented sampling method that enables stable end-to-end value-gradient optimization for flow-matching policies. VINE reconstructs a new interpolation state at each denoising step, creating a stable differentiable path for value-gradient propagation. The method preserves the expressiveness and iterative generation of flow-matching policies and achieves stable policy improvement. VINE outperforms state-of-the-art RL methods on the OGBench offline RL benchmark and real-world robotic manipulation tasks.
Why it matters: This work addresses a key source of instability in training expressive generative control policies for reinforcement learning, enabling more robust and scalable robot learning.
Jul 14, 2026
Research→Official→arXiv Robotics
A new preprint demonstrates that object-centric slot representations, such as SPOT (DINO ViT-B/16 + Slot Attention), substantially improve robotic manipulation success rates in imitation learning tasks compared to dense global features or patch grids. On the ManiSkill3 PickCube-v1 benchmark, a frozen SPOT encoder achieved a 55% success rate, outperforming a dense baseline by 22.4 percentage points, without increasing model capacity or requiring encoder fine-tuning. Further gains were observed by adding explicit spatial goal information and higher-resolution rendering.
Why it matters: This work highlights that structured object-centric representations can meaningfully enhance visuomotor imitation learning for robotic manipulation, offering a practical advance without increasing model complexity.
Jul 14, 2026
Research→Official→arXiv Multiagent Systems
Researchers introduce CoRL-MPPI, a method that integrates Cooperative Reinforcement Learning with Model Predictive Path Integral (MPPI) control for decentralized multi-robot collision avoidance. By training a neural network policy to guide MPPI sampling, the approach improves navigation efficiency and safety while maintaining the theoretical guarantees of MPPI. Experimental results show that CoRL-MPPI outperforms both classical and learning-based baselines in dense, dynamic environments.
Why it matters: This work demonstrates a significant advance in scalable, safe, and efficient multi-robot navigation by combining learning-based behaviors with provably-safe control methods.
Jul 14, 2026
Research→Official→arXiv Robotics
Researchers introduce Multi-Robot Rolling Diffusion (MRRD), a framework designed for real-time, long-horizon navigation of robot teams in crowded human environments. MRRD integrates rolling-horizon planning, parallelized diffusion inference, and conflict-based search to generate socially aware, human-like paths while resolving inter-robot collisions. Experiments demonstrate that MRRD scales to 15 robots in real-time and significantly outperforms existing baselines in both safety and mission success rates.
Why it matters: This work represents a notable advance in multi-robot coordination for crowded, human-shared spaces, supporting safer and more efficient deployment of robot teams in real-world environments.
Jul 14, 2026