Live intelligence

Latest AI updates

11 updates
Research Editor selected

AeroRSI: Recursive Self-Improvement for Aerial Vision-Language Navigation

AeroRSI is an arXiv paper presenting a recursive self-improvement framework for UAV vision-language-action navigation policies. It allows a failing policy to generate and validate repairs by continuing flight from failure states, then distills reliable repairs while preserving behavior elsewhere. On TravelUAV, three rounds are reported to raise a supervised 2B-parameter policy's success rate from 46.4% to 67.5%, halve collisions, and surpass the previous state of the art by 10.6 points without extra human labels or inference cost. Code and models are linked in the abstract.

Why it matters: It offers a concrete method and reported benchmark gains for improving closed-loop aerial vision-language-action navigation without additional human labels.

Vision-language-action modelsUAV navigationRecursive self-improvement
Research

Single VLA checkpoint tackles 100 household tasks in BEHAVIOR 2026

The team describes its 2026 BEHAVIOR Challenge solution: one vision-language-action checkpoint for all 100 long-horizon household tasks in OmniGibson, built on its 2025 entry based on pi_0.5. It placed 9th, reporting Q 0.371 on the 50 tasks shared with 2025 versus 0.260 previously, and 0.392 across all 100 tasks. The work focused on longer training, a 21-dimensional action space, variance-reduced multi-sample flow matching, and cloud evaluation infrastructure. For 14 tasks, the system routes control to a GPT-6 Astra coding agent, which achieved mean Q 0.639 versus 0.108 for the learned policy.

BEHAVIOR Challengevision-language-action modelshousehold robotics
Research

Be My Eye: Learning to See Beyond a Robot's Own Perspective

The paper proposes a task-oriented active context sharing framework for multi-robot collaboration under partial observability, arguing that robots should selectively encode task-relevant information rather than passively transmit raw observations or generic visual features. Onboard observations and task instructions are jointly compressed into compact context representations guided by task-oriented future prediction and soft task-progress correspondence learning, then integrated into a pretrained Vision-Language-Action policy for bidirectional context exchange. Controlled experiments report substantial performance gains under constrained communication budgets, and three real-world dual-robot experiments validate the framework.

multi-robot collaborationvision-language-action modelsrobot perception
Research

HiMEM brings hippocampus-inspired memory consolidation to vision-language-action models

Researchers introduce HiMEM, a memory framework for Vision-Language-Action models operating in complex 3D scenes. It consolidates episodic experiences into compact semantic memories, prunes redundant episodic memories, and composes multiple memories to improve cross-scenario generalization. Experiments on LIBERO and LIBERO-Plus show gains across VLA backbones, including raising pi_0's average success rate on LIBERO from 91.88% to 98.20%.

Vision-Language-Action modelsMemory mechanismsRobot learning
Research Editor selected

On Agentic Manipulation Policies

A new arXiv paper analyzes agentic manipulation policies (AMPs), where a generalist AI model directly commands a robot manipulator and adapts actions as tasks unfold. The authors present AMP1, a reference harness that runs unchanged across single-arm, bimanual, and mobile manipulation without demonstrations or task-specific training. AMP1 reports 85.6% zero-shot success on RoboLab-120 versus 49.2% and 42.9% for leading VLA and WAM policies, and succeeds in 104 of 108 real-world trials across twelve tasks and three robot instances. Adding a persistent memory store lowers failure rates on RoboDojo bimanual tasks by 4.6x relative to zero-shot deployment.

Why it matters: It provides concrete benchmark and real-world evidence that generalist agentic control can outperform specialist robot manipulation policies.

robot manipulationagentic AIvision-language-action models
Research

VLA-Tug: Lightweight Latent Probe Improves Early Failure Detection in Vision-Language-Action Models

Researchers propose VLA-Tug, a lightweight probe over latent layers of Vision-Language-Action models that improves early failure detection and bounds missed-detection and false-alarm rates with minimal inference overhead. It increases early detection by up to 30% over the strongest baseline and maintains valid bounds in up to 86% of runs under task shift. VLA-Tug also predicts failure modes, boosting top-3 accuracy by up to 44% over a majority baseline, and includes a dataset of 3,180 annotated manipulation episodes.

Vision-Language-Action modelsfailure predictionrobotics safety
Research Editor selected

Asynchronous Robot Policy decouples VLM reasoning from fast action control

ARP is an architectural framework that separates high-capacity VLM semantic reasoning from lightweight action generation using asynchronous processes at different frequencies. It updates a semantic memory at low frequency while a Vision-Action Expert generates actions at high frequency, trained to handle staleness and execution delay. On Franka tasks, ARP improves success rate by 18%, reduces execution time by 20%, and achieves 3-5x baseline performance in dynamic simulations, while allowing the VLM to run at 10x lower frequency without significant loss.

Why it matters: This is a concrete architectural advance addressing the latency-reactivity tradeoff in VLA robots, with clear quantitative gains and practical efficiency benefits.

robot manipulationvision-language-action modelsasynchronous processing
Research Editor selected

VLA Policies Do Not Know When Not to Act: Unintended Task Execution and Start Suppression

A new arXiv study examines whether vision-language-action policies refrain from acting when instructions are missing, meaningless, infeasible, or refer to the wrong task. Across six checkpoints in LIBERO-Goal simulation, an empty instruction still caused movement in every episode, and five checkpoints often completed a task the instruction had not specified. Three suppression approaches were tested: an instruction prefix failed across all six checkpoints, an action-chunk gate held the robot on non-task inputs in five of six, and fine-tuning prevented task completion on non-task inputs but did not reliably keep the robot still or handle infeasible instructions.

Why it matters: It provides concrete evidence that current VLA policies can start and execute tasks without valid instructions, an important safety issue for embodied AI deployment.

vision-language-action modelsrobot safetyinstruction following
Research Editor selected

BURGER benchmark tests VLA generalization through robotic drawing

Researchers introduced BURGER, a language-conditioned robotic drawing benchmark with a prompt-driven pipeline for generating continuous drawing trajectories in simulation and the real world. Using 28,200 simulated demonstrations across 282 drawing tasks, they evaluated five representative vision-language-action models on data scaling, language grounding, and multimodal behavior. The benchmark revealed distinct failure patterns and indicated that some VLA models benefit from greater task diversity and demonstration scale and can generalize to unseen compositions. A trained policy was also deployed on a Franka Panda arm to draw reactively under human intervention.

Why it matters: It provides a structured benchmark and empirical analysis for measuring how well vision-language-action models generalize across language-conditioned robotic tasks.

vision-language-action modelsrobotic generalizationbenchmarks
Models Editor selected

Kyutai Releases REKI, a 2.4B Vision-Language-Action Model with Cached Register Memory

Kyutai, with Inria and École des Ponts, has released REKI, a 2.4B-parameter vision-language-action model for robot manipulation that was pretrained only on public data. It pairs a Qwen3.5-2B backbone with a 200M-parameter flow-matching action expert, and adds memory through 32 learned register tokens cached over the last three frames rather than feeding past frames into the VLM. The authors report 90.4% on LIBERO-Plus and 65.8% on RoboTwin Clean2Rand, with plain-language controls cutting arm osc

Why it matters: It offers a concrete, openly released VLA and a cheap memory design that robotics practitioners can inspect, reproduce and finetune.

Kyutaivision-language-action modelsrobot manipulation
Opinion

MIT Technology Review Examines the Gap Between Humanoid Robot Hype and Real AI Progress

MIT Technology Review reports that while Elon Musk and others predict humanoid robots will soon transform labor and daily life, many robotics researchers remain skeptical. The piece details current systems like Google DeepMind's Gemini Robotics running on ALOHA 2 and Physical Intelligence's π0 series, which can perform some general tasks but fail outside their training data. It contrasts flashy, often teleoperated demos with lab progress toward vision-language-action models and world models. Res

humanoid roboticsvision-language-action modelsworld models