Stretch Community News - September 2023
Welcome to the Hello Robot monthly community update!
Last month, Stretch got to meet hundreds of you at IROS in Detroit, a great event that concluded with research from Meta and Georgia Tech presented live on Stretches at the Demonstration Expo. We hope to see more of you at CORL in Atlanta next month!
Read on for details of more recent work being done with Stretch. If you’d like to see your work featured in a future newsletter or on the Stretch Community Repo, please let us know!
Stretch’s potential to help individuals with disabilities was featured in this month’s IEEE Spectrum. Highlighting recent work by Vinitha Ranganeni (University of Washington/Hello Robot), the article describes Hello Robot’s ongoing research with Robots For Humanity’s Henry Evans and displays Stretch’s ability to become a platform that enables independence and empowerment.
Spatial Language Attention Policies (SLAP) allows language-guided manipulation to extend to mobile manipulators! Designed by researchers at Meta AI, CMU, and SUNY Binghamton, SLAP uses a point-cloud based ”tokenization” to handle spatial awareness in a way that is independent of the robot’s exact embodiment.
Georgia Tech students unveiled ForceSight - a vision transformer that produces visual force goals from only an RGBD image and a text prompt, empowering mobile manipulators like Stretch to plan and execute contact-rich tasks.
At Cornell, Prof. Sanjiban Choudhury’s lab is working towards superhuman imitation learning - training Stretch to outperform a set of imperfect training data on household tasks like drawer opening.
For a robot to anticipate user needs without explicit requests, it must understand routines and predict consistent and inconsistent behaviors. Georgia Tech researchers propose a new model for longitudinal proactive assistance, in which the robot learns user behavior patterns from observations of household tasks, and then provides assistance by fetching objects prior to being asked. Video Link
