Topics
Explore 13 articles on Embodied AI models from How Robots Learn.
A demonstration is compiled into coordinates describing object motion, contact locations and the desired final scene. It holds that effect program fixed while generating an execution from the live scene, so people can edit the intended outcome and the robot can re-solve the task after changes.
Published
Uni-VLaT adds a distributed tactile pathway to pretrained VLA policies and trains it to predict future touch, proprioceptive, and visual representations alongside actions. Its authors report a 75% average success rate across five real-robot tasks, while the single-preprint evidence does not provide independent confirmation.
Published
A preprint proposes encoding scene contours, goals, action targets and future states as spatial sequences, then training one autoregressive Transformer for both action generation and action-conditioned forecasting. Its simulation and real-robot results are author-reported.
Written
Both occupy a shared metric frame, allowing one model interface to produce executable actions and action-conditioned forecasts of physical change.
Written
PLaW-VLA predicts task-relevant futures in a pretrained latent representation and conditions action generation on them through structured causal attention. Its authors report benchmark gains and lower latency than generative world-action modeling.
Written
Compare a skill selector with an action predictor, using the interfaces described by the SayCan and OpenVLA authors.
Written
The I-BFM authors describe a policy that learns shared humanoid, object and contact dynamics, uses interaction and goal targets across different time scales, and follows latent commands associated with downstream rewards. They report 94.3% nominal success on Carry and 89.3% after robot falls, alongside experiments on a Unitree G1.
Written
ROMA is a proposed active-perception system that identifies an evidence gap, chooses what object to test and how to test it, then incorporates newly collected multisensory feedback into its reasoning loop.
Written
CompVLA pairs motion prediction with a Compliance Expert that produces time-varying stiffness and virtual displacement profiles for geometric impedance control. Its reported comparative performance remains a claim by the preprint authors.
Written
The authors report an 81.2% success rate on the in-domain SimplerEnv WidowX benchmark.
Written
PRISM uses video-to-video generation to amplify a few human-object interaction recordings into hundreds of varied examples, then reconstructs and retargets their motions into physically plausible trajectories. Its authors report category-level generalization and deployment on a real humanoid without real-world fine-tuning.
Written
SLIP-VLA creates future latent representations with one denoising update, aligns them with geometric and semantic features, and connects predicted transitions to robot actions. Its performance and 12 ms latency are claims reported by the preprint’s authors.
Written
MorFiC's conditional quadruped transfer, the contrast between speed and survival, physical Go1 and Go2 trials, and the limits shown by B1.
Written