§ Robotics

Baidu's Apollo simulator uses AI to generate rare autonomous vehicle edge cases for robotaxi training.

Cycle WAM enforces cycle consistency between predicted robot actions and video consequences via pretrained video priors to improve world action model training.

CLARA framework grounds ambiguous robot/LVLM instructions via cognitively-driven clarification, enabling systems to disambiguate high-level requests before execution.

CANDO: noise-space residual RL method for efficiently adapting frozen pretrained robot control policies.

DiMuRe framework for robotic agents uses diagnosis-guided recovery with MLLMs to select task reuse or predefined skills when disturbances occur during execution.

Nvidia showcases AI and robotics partnerships at Seoul event with Samsung, SK Hynix, LG, Doosan, SK Telecom, Naver, Coupang.

Visual adversarial perturbations can alter physical actions in reasoning VLAs while preserving identical text rationales; affects Alpamayo, InstructVLA.

Apple plans $1,000 home robot by 2027 for FaceTime calls and light control, led by John Ternus.

Tutor raises $75M Series B (a16z-led) for frontier robot foundation models and warehouse automation, reaching $117M total funding.

AI agent autonomously performed minimally invasive brain surgery on a mouse with robotic precision, advancing surgical automation.

Jinpao Precision Industry runs robotics competition in Thailand to develop AI and manufacturing talent.

Foxconn expands medical robotics and hospital AI workflows via ecosystem partnerships and edge-cloud integration.

LG Electronics, LG Innotek, and Hyundai Motor advance Nvidia robotics partnerships with data factory operations and autonomous driving infrastructure.

Safety4DM uses vision-language models to extract risk entities from road observations for vehicle-road-cloud cooperative autonomous driving.

TriPhysBench evaluates physical intelligence in embodied agents across 410 questions integrating perception, knowledge, and reasoning.

Neural control barrier functions enable safe latent-space robotic motion planning without explicit collision classifiers.

Team Friday's 19M-parameter policy achieves 0.947 score on MetaFine fine-grained robot manipulation challenge.

SAE steering via per-token feature replacement can redirect robot foundation model π₀.₅ task goals; layer-0 BatchTopK most effective.

X-World 1.6 adds finer traffic control, spatial consistency, and Virtual City scenario construction to generative driving simulator.

Mechanical controller limits explain imitation learning collapse beyond 2–3× demonstration speed in robotics.

Reflex addresses robot policy inference latency via action-chunk caching and denoising-step optimization for multi-robot control.

ID-WM hierarchical world models decouple state reachability from control to improve long-horizon embodied agent planning with reduced cumulative error.

RoboArchitect automates robotic agent design via coding models that iteratively revise control flow around frozen policies from rollout observation.

Sensorium: open-source Python library for recording, processing, and curating multimodal surgical robot data via ROS middleware.

∇Mimic uses differentiable simulation for humanoid control, achieving order-of-magnitude sample efficiency gains over PPO.

VLA policies unexpectedly execute actions on malformed inputs; proposes start suppression mechanism to prevent unintended task execution.

FOCUS applies optimal transport alignment for imitation learning from observations in contact-rich robotic tasks without temporal correspondence.

Honda targets Level 4 autonomous driving by 2027 and AI-based map replacement by 2030 for Asian markets.

HHTD 2026 exhibition showcases robotics and AI infrastructure; participants include Foxconn, KUKA, ABB, Yaskawa, TDK, Hiwin, TM Robot, TECO.

Nvidia projects 69% market share in AI robotics chips by 2026, driven by software ecosystem advantage over Intel, Qualcomm, AMD.

Gym-Anything for Robotics uses coding agents to automatically generate realistic physics simulations for embodied AI training at scale.

BURGER benchmark evaluates vision-language-action model generalization via scalable prompt-driven robotic drawing in simulation and hardware.

Value-of-information framework for efficient memory retrieval and credit assignment in RL with learned world models for latency-constrained robotics.

CoC-CFG: classifier-free guidance method calibrating reasoning-trace influence in VLA policies via learned step-level action-value weighting.

LiDAR-SHIELD provides Dockerized adversarial robustness testing framework for 3D LiDAR perception models in autonomous driving.

AegisFlow enables training-free path-safe guidance for diffusion and flow matching models via non-myopic constraint satisfaction.

Kairos system orchestrates multiple PDDL planning engines by task semantics for real-time robotic control.

Movement-primitive action chunking reduces jitter in robot imitation learning via diffusion and vision-language models.

UMI-Live is an open-source bimanual manipulation interface with integrated sensing for in-the-wild robot learning data collection without external calibration.

WorldFlow jointly predicts video frames and 3D object trajectories for embodied AI planning via mixture-of-tokens pretraining.

Unitree Robotics reports 70% revenue surge to 665M yuan in Q3 but net profit falls 17% amid R&D investment.

CuraEval benchmarks demonstration selection strategies for robot imitation learning on mixed-skill robomimic data.

Vision-Language-Action models trained on 1616 underwater manipulation episodes for robotic task generalization.

Study reveals five joint-space interfaces and filtering properties in robot learning mask implementation details affecting attribution.

ThinkWorld framework grounds VLA robotic learning in world dynamics and RL imagination, scaling beyond action-labeled demos.

Constraint-Referenced Strategy metric evaluates robot manipulation by separating progress along object joints from wasted motion.

Adaptive Residual Correction uses sparse human feedback to guide RL refinement of imitation policies on localized failure regions.

Vision-language models struggle to diagnose bimanual tool failures without access to low-level sensing channels beyond vision.

RL controller for fixed-wing drone interception of high-altitude targets; military application with limited commercial relevance.

300+ Boston Dynamics employees oppose sale of Spot robot dogs to ICE amid ethical concerns over law enforcement deployment.