Prithwish Dan

MS @PortalCornell, Robot Learning

[Kushal @kushalk_]

Play2Perfect learns contact-rich, precise assembly skills like screwing.

The best part is watching the rollouts slowed down: the policy makes tiny corrections and recoveries that are essential for fast, dexterous manipulation.

Check out Tyler’s thread for more details 👇

[Kushal @kushalk_]

Let humans be humans, and let robots be robots. Our paper at #ICRA2026, X-Diffusion views human data as noisy counterparts of robot data.

Building on the Ambient Diffusion framework, we selectively integrate human data in the high-noise timesteps of diffusion policy training 👇

[Tyler Lum @tylerlum23]

🤖 How can we teach dexterous robots to perform precise, contact-rich assembly?

Introducing Play2Perfect: first learn to play with objects, then perfect the policy for tight insertion, multi-part assembly, and screwing.

Sound on! 🔊

🧵👇

[Maximus Pace @maxpace1_]

Teaching a robot shouldn't require humans to act like robots.

Human demonstrations contain valuable signal for robot manipulation, but they aren’t directly transferable to robots.

X-Diffusion learns from noisy human demonstrations while staying within the robot’s capabilities.

[Maximus Pace @maxpace1_]

Human videos 👩 are noisy and break standard imitation learning.

X-Diffusion selectively trains on noisy human data to learn skills while maintaining robot feasibility 🦾

Come check out our poster at #ICRA26
Interactive Session 6 − ThI2I Hall C - 3-4:30 p.m.

[Tenny Yin @tennyyin]

🎮 Can we learn interactive world models from letting robots “play”?

➡️ Introducing ✨PlayWorld: a framework for training high-fidelity video world models from large-scale autonomous play experience that enables:

→ Accurate dynamics prediction
→ Reliable policy evaluation
→ RL fine-tuning entirely inside the world model

🌐https://t.co/Kpd2DoveXc

[Kushal @kushalk_]

🤖 Can a single robot policy manipulate diverse tools without ever seeing them before?

Introducing SimToolReal 🔨 : a generalist dexterous manipulation policy that transfers zero-shot sim→real to unseen tools + unseen tasks

All videos are 1x speed (60 Hz control) 🧵👇

[Yuki Wang @YukiWang_hw]

Using OT to define rewards for imitating video demos is popular, but it breaks down when demos are temporally misaligned—a frequent challenge in practice.

We present ORCA at #ICML2025 , which defines rewards by aligning sequences, rather than matching individual frames via OT.

[Stone Tao @Stone_Tao]

X-Sim, best Paper (Runner-Up) at the EgoAct Workshop @ RSS.

Real2sim2real with pure image based simulation training and reward extracted from human videos

[Kushal @kushalk_]

Huge thanks to my co-lead @prithwish_dan, collaborators Angela Chao, Edward Duan & Maximus Pace, and co-advisors @weichiuma @sanjibac!

Thrilled that our X-Sim paper received Best Paper (Runner-Up) at the EgoAct Workshop @RoboticsSciSys — winning a cool pair of @Meta Ray-Bans! 😎

[Kushal @kushalk_]

Teleoperation is slow, expensive, and difficult to scale. So how can we train our robots instead?

Introducing X-Sim: a real-to-sim-to-real framework that trains image-based policies 1) learned entirely in simulation 2) using rewards from human videos.

https://t.co/5yt2iTFYF4

[Gokul Swamy @g_k_swamy]

Say ahoy to 𝚂𝙰𝙸𝙻𝙾𝚁⛵: a new paradigm of *learning to search* from demonstrations, enabling test-time reasoning about how to recover from mistakes w/o any additional human feedback! 𝚂𝙰𝙸𝙻𝙾𝚁 ⛵ out-performs Diffusion Policies trained via behavioral cloning on 5-10x data!

[Sanjiban Choudhury @sanjibac]

A core challenge we face when training robots from human videos (e.g., MotionTrack) is that the human basically has to move like a robot for it to work...

@kushalk_ and @prithwish_dan have gone deep on this over the past year—first of many exciting papers that will solve this!

[Kushal @kushalk_]

We’re incredibly proud to see our work featured by the Cornell Chronicle!

This work was co-led with the amazing @prithwish_dan and advised by @sanjibac

Read the full story here -
https://t.co/q1BtLXMA0A

[Kushal @kushalk_]

We would love to train our robots with human videos. But humans move very differently from robots! How do we bridge this divide?

Check out our work at #ICRA2025 “One-Shot Imitation under Mismatched Execution” on specifying tasks to robots via a prompt human video 🧵

[Sanjiban Choudhury @sanjibac]

🚀 Come checkout the *four* papers at #CoRL2024 from @PortalCornell ! Each explores a unique angle on how robots can learn effectively with humans:
- Learning by asking (APRICOT)
- Learning by watching (RHyME, Time Your Rewards)
- Learning to collaborate (MOSAIC)

Details: 🧵👇

[Sanjiban Choudhury @sanjibac]

How can we enable LLMs to actively clarify ambiguous task specifications by gathering information from humans?

Check out APRICOT at #CoRL2024! APRICOT combines LLMs, which propose diverse questions, with Bayesian Active Learning, which selects the most informative one to ask.

[GuilderlandAthletics @GoDutchAthletix]

Congratulations to graduating senior Sean O'Brien who will be attending University of Buffalo to run cross country and track! Sean is planning on majoring in Engineering! Good luck Sean!

[GuilderlandAthletics @GoDutchAthletix]

Congratulations to graduating senior Sheridan Dillon who will be attending Fairfield University to play baseball! Sheridan is planning on majoring in Business! Good luck Sheridan!

[Uzi London 🌎☄️💕® @LILUZIVERT]

Sasuke 😈🔥