World Models

PriorEye: Geospatial Visual Priors for Autonomous Driving

PriorEye introduces geospatial visual priors for end-to-end autonomous driving, overcoming the limitation of relying solely on instantaneous sensor observations.

3D HAMSTER: 3D End-Effector Trajectory from RGB-D

3D HAMSTER predicts 3D end-effector trajectories directly from RGB-D, unifying the planner and point-cloud policy in the same coordinate frame.

Beyond Imitation: Self-Improving Robot Policies via Off-Policy Q-Planning

Q-Planning freezes a large behaviour-cloning policy and trains an approximately one-billion-parameter off-policy Q function. At inference, Q performs single-step weighted averaging over BC action-chunk draws; during online self-improvement, both successful and failed rollouts update only Q. Ten iterations raise mean LIBERO success from 92.1% to 97.6%, while two bimanual real-robot tasks improve from 40%/25% to 90%/80%.

Varun Giridhar, Anant Khandelwal, Jeremy A. Collins

DECOWAM: Decoupled Whole-Body World-Action Model for Legged Mobile Manipulation

DECOWAM adapts a frozen FastWAM video-action backbone to legged mobile manipulation via decoupled interfaces — an action-equivalent future bottleneck, adversarial base/arm factorization, and ego-motion-aware video conditioning — cutting Stage-2 trainable parameters 232x while leading real-robot deployment at 58.2% success.

Siyuan Ma, Boshi Zhang, Yutian Zhang

ADEPT: Sim Pre-training for Zero-Shot Visuo-Tactile Dexterity

ADEPT introduces a pre-training and post-training paradigm: robot hands acquire general dexterity via reinforcement learning entirely in simulation, then zero-shot deploy visuo-tactile policies to the real world, reusing learned dexterity to pick up downstream tasks much faster.

Hydra-0: Action Flow for Generalist World Modeling and Control

Hydra-0 introduces a unified action language bridging physics-based and world-model-based simulation, enabling end-to-end connectivity across human-robot interaction, robot control, and simulation.

Build Manipulation Policies with NVIDIA Cosmos 3 Edge

NVIDIA walks through preparing Cosmos 3-DROID data, post-training a 4B model, and running a manipulation policy directly on Jetson Thor with receding-horizon control, no data-center GPU in the loop.

EATR-Stereo: Embodiment-Aware Token Routing of Paired Stereo Evidence for Humanoid Vision-Language-Action Control

Long-horizon humanoid vision–language–action (VLA) control with head-mounted stereo cameras requires visual interfaces that can exploit complementary views while maintaining compatibility with pretrained representations. Existing interfaces often discard complementary stereo evidence or fuse additional observations without preserving the native primary-view pathway and adapting auxiliary information to robot embodiment. We present EATR-Stereo, an embodiment-aware token-routing framework that retains primary-view tokens and constructs primary-aligned Cross-View Auxiliary Tokens (CVATs) by querying the synchronized auxiliary-view token sequence. A body-segmented proprioceptive encoder further conditions token-wise auxiliary usage on robot configuration history, enabling selective incorporation of stereo evidence during action generation. The routed auxiliary stream augments the language and primary-visual context of a pretrained VLA while keeping its vision–language model frozen. On a 33-DoF physical humanoid with a 37-D proprioceptive state, we evaluate nine configurations in over-100-s search–approach–grasp–place–return tasks. EATR-Stereo achieves 60.0% full-task success, 100.0% grasp success, and 80.0% stage success. Under severe asymmetric occlusion, it improves recovery to 80% compared with 30% for CVAT alone. Ablation studies further show the importance of preserving primary tokens and combining cross-view auxiliary features with structured proprioceptive routing. These results demonstrate that selectively routed paired stereo evidence improves spatial grounding for reliable long-horizon humanoid VLA control.

Songwei Wu, Rui Zhao, Fan Yang

HarnessEval-W: an agentic benchmark for visual world models

HarnessEval turns a benchmark from a static rubric into an executable agent: it interprets context, decomposes a high-level evaluation problem into sub-problems, assembles the right tools and spawns sub-agents to uncover why a model fails, leaving a fully verifiable reasoning trace behind every final score. The first released harness targets visual generative world models, with the harness, skill libraries, evaluation cases and results all public: code, leaderboard, blog and a technical report on arXiv (2608.16859).

Symbiosis Robotics puts a Unitree G1 behind a go-kart wheel

Founded less than 60 days ago, Symbiosis Robotics builds robot brains rather than bodies; its first public demo puts a Unitree G1 driving a go-kart. Founder Ding Pengxiang is a Gen-95 PhD from Zhejiang and Westlake universities.

GigaBrain-WBC-0.5: A Behavior World Model for Robust Whole-Body Control with Environment Interaction

The first Behavior World Model (BWM) for humanoid whole-body control. A causal Transformer jointly predicts next action, state, and latent behavior command distribution, enabling the policy network to model how the environment shapes actions. Automatic terrain-annotation pipeline recovers 3D contact geometry from retargeted motion. At deployment, implausible commands are detected and retracted onto learned behaviors. Achieves highest success rate across all four regimes: 81.3% terrain interaction, 83.1% under implausible commands, 99.3% fall recovery. Unitree G1 checkpoint transfers to Maker L01 robot.

Ziyang Cheng, Tianshu Tang, Jinxin Lan

Introducing S1: In-Context Learning for Robotics — One Video Demonstration Is All It Takes

Skild AI's flagship robotic foundation model S1 is built from the ground up as an in-context learner: show it a video demonstration of a task — even an unseen 10-minute long-horizon one — and it executes with no fine-tuning. On unseen tasks, one demonstration ≈ 380 post-training examples; at 100k pre-training hours S1 reaches 66% success vs 9% for language-prompted VLAs.