World Models

VLK: Learning Humanoid Loco-Manipulation from Synthetic Interactions in Reconstructed Scenes

VLK synthesizes paired vision-language-kinematics supervision inside 3DGS-reconstructed real scenes: it generates navigation and object-interaction trajectories with privileged scene info, renders egocentric views after the fact, and produces 48,000 paired trajectories to train a policy predicting Unitree G1 whole-body motion, enabling sim-to-real perception-based humanoid loco-manipulation.

Yen-Jen Wang, Jiaman Li, Sirui Chen

DynaWM: Dynamics-Aware Distillation with World Model and Momentum Targets for Smooth Locomotion over Continuous Stairs

Recent advances in control have enabled bipedal-wheeled robots to traverse slopes and single-step obstacles, yet long staircase traversal remains challenging as current teacher-student frameworks suffer from weakened dynamics-aware representations and incomplete terrain geometry encoding. To bridge this gap, we propose DynaWM, a dynamics-aware representation learning framework. To enhance terrain encoding capability and enable transparent assessment, we introduce a world model as a regularizer to enforce forward-dynamics awareness, preserving comprehensive terrain geometry while facilitating hierarchical encoding visualization. To stabilize knowledge transfer, we employ a momentum target encoder to provide consistent distillation targets, preventing dimensional collapse from non-stationary teacher updates. Evaluation of the learned representations through Principal Component Analysis (PCA) visualization and quantitative metrics reveals that our encoder hierarchically captures terrain geometry with higher terrain encoding capability, leading to enhanced terrain adaptability and motion smoothness. Experimental results in simulation and real hardware demonstrate that our method achieves superior terrain adaptability and motion smoothness, enabling bipedal-wheeled robots to overcome diverse continuous stairs, as shown in Fig. 1.

Haidong Hou, Zhangguo Yu, Hengbo Qi

EgoEngine turns egocentric human videos into robot demos

EgoEngine converts egocentric human videos into robot demonstrations, producing temporally aligned robot-view video and executable action trajectories for zero-shot dexterous policy learning without real-robot demos.

1X World Model: Simulating the Future to Evaluate Robot Policies

1X shares 1XWM progress: action-controllable video generation predicts future outcomes of robot actions, shifting policy evaluation from physical experiments to simulated forecasting at scale.

Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models

Embodied-R1.5 unifies spatial cognition & reasoning, planning & correction, and embodied pointing & location within a single 8B Qwen3-VL backbone, trained on a 15B-token corpus built by three automated pipelines plus a multi-task balanced RL recipe. A Planner-Grounder-Corrector closed loop lets one model drive the full autonomy stack, completing 10-step long-horizon real-robot tasks like making milk tea with autonomous error recovery; Embodied-R1.5-VLA reaches 92.4% on SimplerEnv Google Robot without large-scale action pretraining, beating pi0.5 by 20+ points. SOTA on 16 of 24 embodied benchmarks; weights, data, code and the EmbodiedEvalKit are fully open-sourced.

Yuan, Yifu, Huang, Yaoting, Yao, Xianze

LEGS: Fine-Tuning Teleop-Free VLAs for Humanoid Loco-manipulation in an Embodied Gaussian Splatting World

Training vision-language-action (VLA) policies for humanoid loco-manipulation is constrained by the high cost and complexity of collecting human teleoperation demonstrations. VLA policies fine-tuned in simulators have, until now, failed to transfer effectively in humanoid loco-manipulation tasks. We present LEGS (Loco-manipulation via Embodied Gaussian Splatting), a hybrid simulator that composites a mesh foreground (robot, objects, props) over a photorealistic 3D Gaussian Splatting (3DGS) background reconstructed from a handheld scene capture. LEGS uses a procedural motion-primitive generator to synthesize labeled demonstrations at scale without human teleoperation, and a deterministic two-stage color calibration to align the rendered 3DGS image to the robot's deployment camera. On a Unitree G1 humanoid robot, across three pick-and-place tasks of increasing whole-body difficulty and three VLA backbones (psi_0, pi_0.5, GR00T N1.6), a policy trained purely on LEGS data matches or exceeds one trained on human teleoperation demos on every experiment. It also outperforms a mesh-only simulation baseline that ablates the effect of the 3DGS background, showing that photorealistic rendering is a key enabler for synthetic data transfer. Humanoid motion is recorded independently of scene appearance in LEGS, allowing the same auto-generated demonstrations to be re-rendered under new backgrounds and object meshes–covering a new scene at more than 15x lower cost than teleoperation–to augment training data for robustness to scene variations. Under combined object-and-scene appearance shift, the policy trained on re-rendered LEGS-AUG data maintains task success while the baseline trained on teleoperation data fails entirely. Our project page is located at https://legsvla.github.io/.

Kim, Hojune, Chen, Timothy, Sun, Jiankai

WorldKV: retrieval and compression for world-model memory

WorldKV from KAIST is a training-free world memory that replaces full-history attention with world retrieval, which picks relevant KV chunks by camera and action correspondence, and world compression, which prunes redundant tokens by key-key similarity. It halves KV VRAM and roughly doubles throughput on long rollouts, and plugs into existing autoregressive video world models.

WorldKV: managing a world model's memory like a cache

Memory in autoregressive video world models is a bind: full KV attention preserves revisit consistency, but memory footprint and attention cost grow linearly with rollout length, passing 200 GB per minute and exceeding a B200. Sliding window is real-time but drifts. WorldKV trains nothing and works purely at inference: it retrieves evicted KV chunks by camera/action similarity and compresses each chunk to half size by key-key cosine similarity, fitting roughly twice the history at roughly twice the throughput of full KV, with no memory training at all.

Jung Yi, Minjae Kim, Paul Hyunbin Cho

Scaling Behavior Foundation Model for Humanoid Robots

Researchers scale the Behavior Foundation Model to fully unleash humanoid robot manipulation potential.

World Action Models (WAMs): Unifying VLA and World Models

Introduces World Action Models (WAMs) to unify predictive state modeling with action generation, addressing VLA limitations. Proposes taxonomy and analyzes data ecosystem.

Siyin Wang, Junhao Shi, Zhaoyang Fu

GRASP: Gradient-Based Planning for World Models

GRASP proposes three innovations making long-horizon planning with world models practical: lifting states for parallel-in-time optimization, stopping brittle state gradients while keeping action gradients, and periodic sync refinement. Achieves leading success rates at long horizons on Push-T.

Early access to NVIDIA Isaac GR00T N1.7 is here 🎉 — an open, commerci...

Early access to NVIDIA Isaac GR00T N1.7 is here 🎉 — an open, commercially licensed vision-language-action foundation model for humanoid robots, built for real-world deployment. 🤗 Read the @huggingfac