Pollen Robotics

Last week's efforts paid off! With #LeRobot's team from Hugging Face, Reachy performed autonomous tasks on request. A huge milestone open-source embodied AI. More on LeRobot here: https://lnkd.in/dSptBZ_u  And join our discord to follow progress on Reachy 2: https://lnkd.in/dXPBXJaG

🚀 Tactile Gripper: A Playground for Tactile Sensor Experiments At Pollen Robotics, we have been developing tools to test and integrate tactile sensors into robotic grippers. Our goal is to make it easier for researchers and developers to experiment with different tactile sensing solutions for real-world manipulation tasks. 💡 What we built We developed an open-source Rust-based framework to set up new sensors, collect data, and run experiments efficiently. The first sensor we integrated is the Anyskin, a flexible, plug-and-play skin sensor that detects touch using changes in magnetic fields. Shoutout to Raunaq Bhirangi and the team behind Anyskin for this amazing work! 🔬 What we achieved - We mounted Anyskin on our gripper and tested it in real-world grasping tasks. - By implementing a simple threshold mechanism, we improved grasping of soft objects. - We collected a dataset of slip/no-slip scenarios, now available on HuggingFace https://lnkd.in/eEP6YZQN. - Using this data, we trained a real-time slip detection model that automatically prevents object slippage, also available on HuggingFace https://lnkd.in/eVpsMygC. - With Rerun, we built a visualisation tool to inspect sensor data sequences. 🔗 Open-source tools Our work is open-source, and we'd love the community to contribute! - Codebase on GitHub https://lnkd.in/eSVX-7gJ - Rust tools for gripper control and data visualisation - Rerun integration for sensor data playback 🤝 How you can help We want to scale this project as a community effort! If you are working on tactile sensors, robotics grasping, or AI-driven manipulation, we’d love your feedback, contributions, and ideas. 📩 Reach out to Fabien Danieau and Jonathan Grizou or drop a comment below! Let’s make tactile sensing in robotics more accessible and powerful! #Robotics #TactileSensing #MachineLearning #Grippers #OpenSource #RustLang #Anyskin

Teleoperation without VR 🤓 During the Hackathon, Coralie Deplanne developed a solution for teleoperating Reachy 2 with Lerobot's SO100 arm: it's now possible to control the two arms and the head, with just one arm and a keyboard. Next step: integrate haptic feedback to feel force applied to the robot's arm, and maybe print a second arm to create an arm wrestling competition with Reachy one day... 🥊

We connected GPT-4 to Reachy, and suddenly... it developed a kind of emotional intelligence! The LLM listens to speech and chooses which emotions Reachy should express. We’ll dive into the technical details soon, but in the meantime—there’s been a crime at Pollen Robotics 🚨 Mini Duck, Antoine Pirrone’s project, was found completely unscrewed in the workshop… 🕵️♂️🔍 Anne-Charlotte Passanisi and Remi Fabre stepped in as investigators, using Reachy’s emotional cues to aid their sleuthing. Will our apprentice Sherlock Holmes crack the case? Stay tuned!

Waking up in the Stardew Valley 💤 With expressive movements and smooth articulation, Reachy 2 is designed for AI research and human-robot interaction. But today, it’s just trying to get out of "sleep mode"! 💡 What other human-like actions should we make Reachy 2 do next? Drop your ideas in the comments! 👇 Try it yourself on simulation or on your Reachy 2: https://lnkd.in/edUuJP_q Demo made by Simon Le Goff 🫶🏼

Robot vs Junk Food! Claire Houziel gave Reachy 2 a challenge: sorting healthy and unhealthy foods—with no AI training! Just real-time object detection using pollen_vision and our self-development kit. This experiment helped us test new R&D tools while creating a tutorial for Reachy users. No pre-learned data, just pure decision-making in action! Would you trust a robot to pick your meals? 👉🏼 Try a simple version of this demo : https://lnkd.in/eB2BE5Vk 👉🏼 Github pollen vision : https://lnkd.in/gqP-qFnk

Antun Skuric gave it all to make gravity compensation possible on both arms of Reachy 2!

In order to make the Reachy 2 simulation simple, lightweight, easy to install and use, we’ve worked on making it available on Mujoco. Our goal is to integrate it into Reachy 2's real software stack. Antoine Pirrone, Steve N'Guyen and Clément are working on it, and soon, behavior simulation for the robot will be possible! Plus, you can try it yourself here: https://lnkd.in/ewvpzKXa

Check out Haixuan Xavier Tao's amazing project using Reachy 2 for an object grasping mission! Since it's open source, you can find the code here : https://lnkd.in/enTAcuq6

♟️ Reachess Because chess isn't just played by our team at Pollen: Reachy 2 already proved its chess precision against Remi Fabre on video. For a few months, the Pollen Robotics team played against each other in a chess tournament, where novices and experts put their skills to the test. Now we just have to coach Marcel ... 🐶