
Spot dressed up for Halloween!🎃 It's on a mission for its favorite 'candy'! 🔋 But two 'ghosts' were blocking the path… A fun demo of our new paper on how robots can intelligently 'make way' on cluttered stairs! (1/4) @CMU_Robotics

Robotics Ph.D. student at Carnegie Mellon!

Spot dressed up for Halloween!🎃 It's on a mission for its favorite 'candy'! 🔋 But two 'ghosts' were blocking the path… A fun demo of our new paper on how robots can intelligently 'make way' on cluttered stairs! (1/4) @CMU_Robotics

Very cool work—force as an explicit interface is a powerful abstraction for loco-manipulation. I’ve been exploring a related direction: occupied-hands humanoid pedipulation, using the feet to interact while carrying. Exciting to see Ctrl-F demonstrate this across hands, feet, and real hardware. Congrats to the team!
🤖 A reference motion tells a humanoid where to move, but leaves the interaction underspecified. We argue force is a missing interface. 💡 Introducing Ctrl-F: Force as an Interface for Humanoid Loco-Manipulation. [1/6] 🌐 forcectrl.github.io 📄 arxiv.org/abs/2611.05507

Sharing this work that I was partially involved with at @multiplylabs We show that using contact information as feedback signal during training can help track grasps even on new, unseen objects. All demos are open loop. Excited to show this work. Project page and video: multiplylabs.github.io/oath/ Paper: arxiv.org/abs/2610.14379
Sharing OATH: Object-Agnostic Tracking for Dexterous Hands. We found that observing the error to reference contact points, a single residual policy can track grasps on completely unseen objects, enabling rapid physics-based retargeting.

In OATH, we introduce a single reference-, object-, geometry-, and task-agnostic policy that achieves physically feasible retargeting by leveraging contact error as a feedback signal. Excited to share our work!
Sharing OATH: Object-Agnostic Tracking for Dexterous Hands. We found that observing the error to reference contact points, a single residual policy can track grasps on completely unseen objects, enabling rapid physics-based retargeting.

🔗 Website & CFP: rethinking-uncertainty.githu… Submission Deadline: August 23, 2026 (AoE) Workshop Date: Sept 27, 2026 Location: Pittsburgh, PA

Organized by: Prasanna Sriganesh (CMU), @Junwon__Seo (CMU), Juyeop Han (MIT), Rosario Scalise (UW), Xiaoyi (Jeremy) Cai (Waymo), Zhiting (May) Mei (Princeton), Sertac Karaman (MIT), @andrea_bajcsy (CMU) | Matthew Travers (CMU)

We have a fantastic lineup of invited speakers: @max_simchowitz (CMU), @dineshjayaraman (UPenn), @zsoltkira (Georgia Tech), Bernadette Bucher (UMich), Haruki Nishimura (TRI), Sumit Jha (DARPA)

This raises important questions: What does uncertainty quantification mean in this new era? We are bringing together experts across machine learning, computer vision, and robotics to chart a roadmap for the next generation of trustworthy, uncertainty-aware robotic systems.

🤔 Uncertainty quantification has long been a core aspect of robotics. However, in the past five years, many robotics components have started to rely on — or be replaced by — large pretrained AI models, such as LLMs, VLMs, VLAs, video prediction models, and world models.

Do we still need Uncertainty Quantification for robotics? What does uncertainty mean in the modern robotics paradigm? 💡 We are organizing a workshop “Rethinking Uncertainty for Modern Robotics Paradigms” @ IROS 2026 Submit your work by Aug 23! 👇 (rethinking-uncertainty.githu…)

See all the videos and details on our new project page: ➡️ prassi.me/clearing-clutter-o… Full paper: 📷 arxiv.org/abs/2509.20516 (4/4) #Robotics #Halloween2025 #AI

A planner then computes safe trajectories, which are executed by a learned pedipulation policy. This tight coupling results in much higher success rates and lets the robot reclassify heavy 'immovable' objects. Here's a look at more technical results: (3/4)

For robots to be useful, avoiding obstacles isn't enough—they must be able to interact with them. But pushing is uncertain, and vision alone is unreliable. Our system uses proprioceptive feedback to build and update a 3D World Model and track objects in real-time. (2/4)

The code for our work on staircase perception is now public. This repository includes implementations from our papers on Bayesian staircase estimation and fast staircase detection. Code: github.com/Prassi07/staircas…
Making robots safer on stairs! 💪 Our new work, just accepted to RA-L, introduces a clever way for robots to perceive the whole staircase, even when parts are hidden from view! 👀 Full paper: ieeexplore.ieee.org/document… @CMU_Robotics (1/4)

But that's not all! We also use this information to segment the stair surfaces and remove clutter, enabling safer navigation on staircases. 🤖 Check out our project website for more info and images! 👉 prassi.me/bayesian-staircase… (4/4)

Bayesian Inference: We use this model with Bayesian inference to intelligently combine noisy sensor readings with prior knowledge about staircase already seen. This combination lets us tackle challenges like occlusions, clutter, limited field-of-view and sensor noise. (3/4)