Rishabh Madan

🤖 Robots are usually taught to avoid contact unless it’s at the end effector. But caring for a person, such as rolling them over or repositioning a limb, requires embracing contact along the entire arm. How do we enable robots to perform such contact-rich tasks? Introducing TACTIC: Tactile and Vision Conditioned Contact-Centric Control for Whole-Arm Manipulation. #RSS2026 🧵👇

Can robots learn contact-rich tool use, like operating a pair of scissors✂️ or turning a screwdriver🪛, from human data? Introducing REGRIND: a minimalist retargeting-guided RL recipe for dexterous manipulation. 🌐 yunhaifeng.com/REGRIND/ 📄 Paper + code on the site! 🧵👇

@TongheZhang01 Maybe he never left ;)

He is so back

I like the design and interesting choice of company name! Very uncommon for an industrial robot to be this humane.

@plopesresearch Thank you!

@madrish02 Great work, full body contact always the way to go!

🤖 Robots are usually taught to avoid contact unless it’s at the end effector. But caring for a person, such as rolling them over or repositioning a limb, requires embracing contact along the entire arm. How do we enable robots to perform such contact-rich tasks? Introducing TACTIC: Tactile and Vision Conditioned Contact-Centric Control for Whole-Arm Manipulation. #RSS2026 🧵👇

@chris_j_paxton Wow I read this at some point in the first/second year of my PhD. Time flies!

For people who think I've been a doubter, my very first deep learning paper back in 2018 was using world models with a latent representation for test-time compute / planning to solve previously unseen problems. not a great paper but I wrote it!

1/ Progress in robot locomotion scaled with simulation, but touch was left behind. Our work, Tactile Genesis, unlocks sim for dexterous manipulation. Using sim to ablate tactile types and models, we can quickly converge to better tactile hardware and force-informed policies.

TACTIC consistently outperforms both model-based and model-free baselines on whole-arm manipulation. The system is validated on real robots with distributed tactile skin across challenging multi-contact tasks like: • Turning over a manikin • Repositioning a person • Goal reaching through a dynamic 3D maze A great example of combining learned world models with analytical contact physics instead of relying on either alone.

Whole-arm manipulation is fundamentally different from grasping. As contacts form, slide, and break across the entire arm, the robot must reason about both motion and interaction forces—not just end-effector poses. TACTIC tackles this with a contact-centric controller that jointly reasons over vision, tactile sensing, and robot kinematics to predict future contact states.

@picocreator Reminded me of something TRI was experimenting with a while ago core77.com/posts/102246/T…

Surprised we have yet to see a robotics startup pitching a household rail system, which skips all the current hardware problems of batteries and legs This lets it focus on the MVP, doing chores (in house), and hands.

@AnhDaDen811 @AI14935242 Yes it is! If you’re interested, our lab has more work that looks into physical robot caregiving at emprise.cs.cornell.edu

@AI14935242 介護ロボットの進化楽しみだね！

介護ロボットかな？

@Shoier__ @AngchenXie Working on it! 🫡

@madrish02 @AngchenXie Have you tried this in a real human being ?

@oonarisseadams Thank you!

@madrish02 So cool!

Proud of @madrish02 and the entire team's work on TACTIC. What I find most compelling about TACTIC is that contact is not treated as just another sensor input or a penalty term, it is built into every layer of the planning stack. In the representation, RGB-D, proprioception, and distributed tactile sensing are fused through a proximity mask that focuses the model on the interaction geometry that actually matters: where contact is happening, and where it may happen next. In sampling, contact Jacobians shape MPPI exploration, biasing candidate actions toward directions that can meaningfully regulate forces and steering unsafe samples away from high-force contacts. In prediction, hybrid rollouts combine learned interaction dynamics with analytical robot kinematics: learn what is hard to model, while preserving the structure we already know. Interestingly, this hybrid variant idea that adds a kinematics-based goal cost during planning also improves planning with other learned world models, including V-JEPA2 on Reach and DINO-WM on Granular. And in scoring, trajectories are evaluated not only for task progress, but also for contact evolution, force regulation, joint limits, and collisions. That contact-centricity across representation, sampling, prediction, and scoring is, to me, the most exciting part of TACTIC. Very proud of the team and excited to see this work at #RSS2026! @EmpriseLab @Cornell_Bowers @Cornell_CS

Whole-arm tactile sensing, a hybrid predictive framework, challenging real-world tasks relevant to pHRI: new work by @madrish02 has many components I find very cool! Check out his work at RSS 2026 this year.

Lastly, this project wouldn't have been possible without the guidance of @TapoBhat, @mazrk7, and Jose Barreiros. PS: Since I won't be there in person, if you want to chat about whole-arm manipulation, tactile sensing, or JEPA for robotics, my DMs are open!

Sadly, I won’t be able to make it to Sydney in person due to visa issues. Huge thanks to my labmate @ZhanxinWu0725 for stepping up to present our work! Catch the talk at #RSS2026 🇦🇺 🗓️ Manipulation-II — Tue Jul 14, 11:50–12:30 📄 Paper, videos & hardware: emprise.cs.cornell.edu/tactic