More details on Meta FAIR's latest advancements in embodied AI: https://go.fb.me/7mrqd6 Researchers at Meta FAIR are releasing several new research artifacts that advance robotics and support our goal of reaching advanced machine intelligence (AMI). These include Meta Sparsh, the first general-purpose encoder for vision-based tactile sensing that works across many tactile sensors and many tasks; Meta Digit 360, an artificial fingertip-based tactile sensor that delivers detailed touch data with human-level precision and touch-sensing; and Meta Digit Plexus, a standardized platform for robotic sensor connections and interactions that enables seamless data collection, control and analysis over a single cable. To make these advancements more accessible for different applications, we’re partnering with @GelSight and @wonikroboticsto develop and commercialize these touch-sensing innovations. Additionally we're releasing PARTNR, a benchmark for Planning And Reasoning Tasks in humaN-Robot collaboration. It’s the largest benchmark of its kind to study and evaluate human-robot collaboration in household activities. ## Additional resources * Meta Sparsh research paper: https://go.fb.me/7wa7sd * Meta Sparsh repo: https://go.fb.me/wkyie0 * Meta Sparsh models on @HuggingFace: https://go.fb.me/ckhrvy * Meta Digit Plexus repo: https://go.fb.me/a8hp7b * Digit 360 research paper: https://go.fb.me/9a6cga * Digit 360 CFP: https://go.fb.me/dgp15i * Next generation of Allegro Hand built with Meta — interest form: https://go.fb.me/i1lm1s * PARTNR research paper: https://go.fb.me/2ta3y4 * PARTNR website: https://go.fb.me/1vy7vh * PARTNR repo: https://go.fb.me/7bbfz6 -- Subscribe: https://www.youtube.com/aiatmeta?sub_confirmation=1 Learn more about our work: https://ai.meta.com Follow us on Twitter: https://twitter.com/aiatmeta Follow us on Facebook: https://www.facebook.com/aiatmeta Connect with us on LinkedIn: https://www.linkedin.com/showcase/aiatmeta/ Meta focuses on bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.

We present how to train a dexterous hand to rotate dozens of different objects, without vision and without touch. From a stable grasp, our policy can manipulate objects of different weights from 5g to 200g, sizes from 4.5 cm to 8cm, different shapes such as spheres, cubes, and one with holes. None of the above objects are seen during training, the policy only sees cylinders in a simulator and transfer to the real-world without finetuning. To learn more about our project, check out our: paper: https://arxiv.org/abs/2210.04887 website: https://haozhi.io/hora/ github: https://github.com/HaozhiQi/hora/ Published at Conference on Robot Learning (CoRL), 2022.

We present rotateit: how to enable a robot hand to rotate many objects over many different axes. Published at Conference on Robot Learning (CoRL), 2023.

https://toruowo.github.io/bimanual-twist We train two anthropomorphic robot hands to twist (off) lids of different articulated objects. The control policy is first learned in simulation with deep reinforcement learning, then zero-shot transferred to a real-world physical setup. A single policy trained with a simplistic bottle-like object generalizes to objects with drastically different physical properties (e.g. shape, size, color, material, mass).