Nvidia's GR00T 2 turns human video into robot actions
Lab success rates ranged from 10 to 80 percent; with a planner, toy assembly reached at least 97 percent of trials.
Stories per day · 30 days
15 RSS
Relatedrobotics21reinforcement-learning11video-generation11embodied-ai8planning7vision-language-action6evaluation5benchmarks4agents3autoregressive3diffusion3distillation3egocentric-video3multimodal3simulation3spatial-reasoning3video-diffusion3benchmark2
156 stories, newest first · one card per story
Stories tagged #world-models
LLM world models pass mechanical checks but miss intended structure
Of 134 valid game-track models, mean structural F1 is 0.366 and only 28 exactly match the reference.
DexForesight lifts dexterous policies with implicit spatial foresight
On 11 DexJoCo tasks success rises from 49.4% to 60.5%; three real-robot tasks reach 58.3% versus 50.0%.
WM-PPO values world-model predictions to improve sparse-reward agents
With a similar number of real steps, success rises from 63.5% to 87.5% on ALFWorld and from 39.3% to 52.9% on ScienceWorld.
PhysOPSD uses executable code as privilege for video generation
Authors say the method beats supervised fine-tuning and reinforcement learning with 3,000 samples while keeping general video quality.
RoboScript learns manipulation tasks in context from one demonstration
Authors report 83.4% mean success on all 50 RoboTwin 2.0 tasks without gradient updates, over 20 points above the strongest trained policy.
ProTEUS proposes a mutation-resolved model for protein evolution
Authors report variants of aldehyde dehydrogenase and levansucrase with more than 10-fold higher activity than wild type.
WARC learns unseen game mechanics from limited test-time exposures
The 35M-parameter model predicts four in five held-out mechanics at 80% cell accuracy; half reach that within about 80 exposures.
WorldFoundry catalogs 270 models for world-model evaluation
The open-source stack also lists 74 benchmarks and 62 metrics, separating execution evidence from scores.
History Forcing improves long-horizon video world models
History Forcing is a distillation framework that strengthens historical supervision for few-step autoregressive video generators, achieving state-of-the-art results.
ID-WM decouples reachability from control in hierarchical world models
ID-WM uses high-level sampling of feasible transitions and low-level inverse dynamics solving to enable scalable long-horizon planning with 38.5% better performance.
CausalWAM separates action causes from scene prediction in world models
CausalWAM adds sparse routing to pretrained WAMs proving action-relevant latents are identifiable up to permutation for improved robotic closed-loop control.
world2vec learns mechanistic world models from observations and actions
Latent dynamics use at least ten times fewer parameters than DreamerV3, TD-MPC2 and LeWM, without reward.
Probes test spatial state in ten visual foundation models
Strong visible-scene readout does not ensure object persistence, future-feature retrieval or reliable action utility.
Ex-Palantir founders raise $17 million for construction AI startup Dream
Andreessen Horowitz led the seed round; Dream says its model is already used on 10 projects in Texas and California.
1
Paper ties RL memory retrieval and credit to value of information
A learned world model estimates that quantity from archived rollouts, defining an optimal retrieval set and a TD-target floor.
L3G-World generates consistent worlds in real time
Latent 3D Gaussian memory grows with scene extent, not stream length, and the authors report 9 FPS generation.
Wizard enables real-time agentic control of game worlds
Wizard uses skeleton and geometry proxies with adaptive masking to allow zero-shot control of characters in novel scenes via an agentic framework.
WorldFlow Jointly Predicts Video and 3D Motion
WorldFlow reduces ADE by 54.2% on SpatialCog set versus baselines using a Mixture-of-Transformers backbone.
ThinkWorld trains robot action reasoning from unlabeled human video
Chain-of-Dynamics tokens let human and robot video supervise the same action-reasoning space.
Studied latent world models fail a faithfulness test
Latent rollout consistency checks whether action-conditioned latent prediction matches the environment transition map.
Dwell lets vision-language agents learn policies in latent space
The framework treats a pretrained VLM as a recurrent latent transition model and reports top trainable results on five visual environments with less imagination compute.
PDRVLA conditions robot actions on physical dynamics without video generation
It reports 99.3% success on LIBERO and 81.5% on SimplerEnv using public data and a frozen world model.
ThinkWorld grounds robot action reasoning in imagined world dynamics
Chain-of-Dynamics tokens let unlabeled human and robot videos supervise the same action space via predicted state transitions.