RT @DJiafei: 1/ 🚀 Announcing #GenPriors — the CoRL 2025 workshop on Generalizable Priors for Robot Manipulation!. 📍 Seoul, Korea 📅 Sat 27 S….

I am very happy and proud today that my student Norman Di Palo (@normandipalo) has just passed his PhD viva!. Norman has had a very creative few years in my lab and I've learned a lot from his insights and curiosity. Congratulations Norman, it has been a pleasure working with

Vitalis and I had a cracking chat about "Instant Policy" with @micoolcho and @chris_j_paxton. Thanks for the invite!. If you want to know how to really get in-context learning working in robotics, @vitalisvos19 and I take you through it in the video below 👇.

RT @RoboPapers: Full episode coming soon!. Geeking out with @vitalisvos19 & @Ed__Johns on Instant Policy: In-Context Imitation Learning via….

A few years ago, humanoids with legs walking around the ICRA exhibition was the new thing. This time, it’s the year of the hands! Tons and tons of humanoid hands! #ICRA2025

At #ICRA2025, I'm about to present:."R+X: Retrieval and Execution from Everyday Human Videos". By using a VLM for retrieval and in-context IL for execution, robots can now learn from just videos of humans!. Led by @geopgs and @normandipalo. See:

RT @normandipalo: we open sourced the code to transform human videos into robot trajectories, so you can train robots with your hands 👐🏻. w….

Tomorrow morning at #ICRA2025, I will be presenting our findings on whether robots can learn dual-arm tasks from just a single demonstration. (Spoiler: they can!). Come along!. This was led by my excellent PhD student Yilong Wang. Paper & videos here:

Code: Paper: Thanks @geopgs for the code, and @normandipalo and @pitvit_ for the collaboration!

In advance of presenting our ICRA 2025 paper tomorrow, we have released our code for "R+X: Retrieval and Execution from Everyday Human Videos". You can now use this code yourself to extract robot actions just from videos of human demonstrations!. See below for more info 👇👇👇

Thank you very much to @Andrey__Kolobov, @shahdhruv_, @AlexBewleyAI, and the rest of the organisers, for delivering an excellent and packed-out workshop!.

Vitalis (@vitalisvos19) and I were really honoured today to win the Best Paper Award at the ICLR 2025 Robot Learning Workshop, for our paper “Instant Policy”. The video below shows Instant Policy in action… A single demonstration is all you need!. See:

This was led by my excellent student Vitalis Vosylius (@vitalisvos19), in the final project of his PhD. To read the paper and see more videos, please visit (7/7)

Beyond just regular imitation learning, we also discovered two intriguing downstream applications:. (1) Cross-embodiment transfer from human-hand demonstrations to robot policies. (2) Zero-shot transfer to language-defined tasks without needing large language-annotated datasets.

One very exciting aspect of Instant Policy is that the network can be trained primarily with "pseudo-demonstrations": arbitrary trajectories with random objects, all in simulation. And we found very promising scaling laws: we can continue to generate these pseudo-demonstrations

The figure below shows our network architecture, which jointly expresses the context (demonstrations, as sequences of observations and actions), the current observation, and the future actions. Observations are point clouds, and actions are relative gripper poses. During

Instant Policy can learn some tasks with one demonstration. But if a task requires more than one demonstration, it can condition on however many demonstrations are required. In this video, we show how the robot continues to improve and generalise its learned behaviour as further

In-context learning is where a trained model accepts examples of a new task (the "context") at its input, and can then make predictions for that same task given a new instance of it, without any further training or weight updates. Achieving this in robotics is very exciting:

Today, we are presenting "Instant Policy" in our #ICLR oral!. Below is a single uncut video: after just one demonstration for a task, the robot has learned that task instantly. So. we've achieved in-context learning in robotics! . See: (1/7) 🧵

This was led by my excellent student Vitalis Vosylius (@vitalisvos19), in the final project of his PhD. To read the paper and see more videos, please visit (7/7)