FAA For Self Driving Cars 🚗 w/ Neil deGrasse Tyson via @YouTube.

@mariah17schrum was the co-first author on MIND MELD.

Check out some of her papers here!.-Robots learning from older adults (.-MIND MELD: suboptimal demos in imitation learning (.-What happens when the robot fails to learn? (.@core_robotics @GTrobotics.

One last PhD hooding to celebrate from 2024. Congrats to Dr. Erin Hedlund-Botti! She earned a best paper award ⁦@HRI_Conference⁩⁩ and ran multiple studies with real users teaching real robots assistive skills, bringing us closer to imitation learning in the wild. ⁦

Hooding** :)

Out some of their latest work, see these papers:.-Diffusion for multi-agent systems: .-Diffusion for robotics .-VLMs for CAD Code Design: .-LLMs for Alignment in Planning .

Delighted to cap this year off holding @Seancye and @Pradyumna_T. Their legacy for the @core_robotics is immense, pioneering work on generative AI (diffusion models and LLMs). Your leadership will be missed. @zoox and Toyota Research Institute are lucky to have you! To check…

@nina_moorman @nakulgopalan Thanks to all of our co-authors, @nina_moorman, Aman Sing, Manisha Natarajan, Erin Hedlund-Botti, .@mariah17schrum, Chuxuan Yang, Lakshmi Seelam, and @nakulgopalan (ASU). @GTrobotics @gtcomputing @ArizonaState @core_robotics.

This work was the culmination of two Robotics: Science and Systems papers, including a best student paper nomination, and feedback from experts in the field of task and motion planning. Congrats to first author @nina_moorman and senior author @nakulgopalan on this accomplishment.

For these and more insights, including the user's personality traits and prior teaching experience, please check out our paper here: .

These results are critical to support the concept of user programming by demonstration, as prior work had not yet shown that users can (1) learn to teach robots with video examples but without expert coaching and (2) then transfer that experience to teach robots in *new* domains!.

Findings:.1️⃣ Only ~35% of users naturally provided useful abstractions when teaching. 2️⃣ Watching an expert demonstration video helped 100% of participants improve their teaching. 3️⃣ Practice across domains improved abstraction quality (left) & time (right), *even w/o coaching*.

We ran 3 user studies where users taught a robot using demonstrations and a novel Task and Motion Planning GUI:.1️⃣ Do people naturally teach robots using abstractions?.2️⃣ What helps users provide better demonstrations?.3️⃣ Can experience teaching transfer to new tasks/domains?

What’s the challenge? Robots need abstractions to generalize tasks. An abstraction breaks a task into reusable, smaller steps (e.g., "pick up an egg" vs. "make an omelet"). Without abstractions, robots struggle to adapt to new or slightly different tasks.

How can everyday users teach robots to perform complex, long-horizon tasks (e.g., meal prep) without relying on robotics experts for help? Our latest study published in #IJRR tackles this challenge, paving the way for more accessible human-robot collaboration. A 🧵.

RT @rohanpaleja27: Starting travel to #NeurIPS2024! . If you are interested in Human-AI collaboration across repeated games, please come fi….

RT @mark_riedl: Reminder that my Department, the School of Interactive Computing is hiring faculty in AI, ML, NLP, Computer vision, human-c….

RT @ZelenskyyUa: @lexfridman Agreed. See you in Ukraine, @lexfridman.

RT @rohanpaleja27: Ever think about how factors like transparency and policy changes between episodes influence human-AI collaboration acro….

A little late, but Happy Halloween from the @core_robotics. My students worked with Prof. Sonia Chernova's lab to get our Spot to hand out candy to the kiddos. #TrickOrTreat #Robotics @BostonDynamics @GTrobotics