PoRTaL social was a blast ๐ Thank you for the amazing turnout! Our robots ๐ค, HAL and DORA, loved showing off their collaborative cooking skills ๐ฉโ๐ณ Check out the demo here: https://t.co/MF8dkagRPw.
We are just getting started ...
We know LLMs can generate code from language, but wouldnโt it be great if they could do it from demonstrations?
We present Demo2Code, a powerful way to generate robot code from demonstrations! #Neurips2023
Come check out our poster#436 on Thurs 10.45 to learn more! ๐งต
If we don't want robots driving YOU out of YOUR home, we better get robots to learn about humans!
At #CORL2023, we present ManiCast, a novel framework that learns cost-aware human forecasts allowing our robot to collaborate safely and legibly alongside a human partner. ๐งต
Excited to share our work on predicting rare but risky events for safe planning around humans. We propose a new way to jointly train forecasting+planning! Awesome collaboration between @kushalk_ and @prithwish_dan
How do we get robots to forecast rare but risky events to be safe around humans? At #IROS2023, we present a brand new perspective on forecasting that helps robots hedge as well as humans. ๐งต
Tuesday afternoon Poster Session: TuBT9.10
Humans are not perfect. But can robots learn from our imperfections, and be a little bit better?
Excited to kick-off our first NSF project with Brian Ziebart from UIC on "superhuman" imitation learning ๐ Stay tuned for new papers and results in the coming months!
Can we improve #AI to help robots respond to complex real-world situations? ๐ฆพ๐ค๐ญ
Scientists in @Cornell_CS are exploring a "superhuman" approach using a series of imperfect demonstrations - with the robot finding the best way to accomplish the task. https://t.co/N57l2xZKYt
Exciting new work from Portal on learning from demonstrations with LLMs! Awesome collaboration from @YukiWang_hw, Gonzalo and @yash_s20!
https://t.co/vYadUR7GIa
Language is a powerful tool for instructing AI, but it isn't always the most intuitive. Often, demonstrating the task and letting AI imitate us is much simpler. Like making that perfect burger! Introducing Demo2Code, that uses LLMs to translate demonstrations to imitative code ๐งต
Looking back this semester, PoRTaL has come a long way! We grew bigger and closer as a team, took on ambitious projects and shared our story with the world. Come share our memories on instagram ...
https://t.co/CJIVoboSYS https://t.co/2NgItt8cYi
DORA uses deep learning to forecast how humans move around it and plans with these learned forecasts. We collect accurate, high-quality motion capture data of humans and robots working together to train our models.
Stay tuned for upcoming papers on how we train DORA!
Kitchens can get crowded and chaotic ๐งโ๐ณ. Which is why are building DORA (DextrOus Robot Apprentice) to cook alongside you, making the process more enjoyable and efficient. Checkout how DORA adapts and plans to work collaboratively with human partners.
https://t.co/YMWjPYRdi8
HAL uses #GPT to turn natural conversations to task code, and visual imitation learning to learn skills like picking up new objects.
Come by our lab social this Monday 5/8, 5 pm, Rhodes 551 to see HAL do this live. We have food ๐ช, we have drinks ๐ฅค, we have robots! ๐ค
Cooking is fun! ๐ฉโ๐ณ But searching for condiments ... not so much. ๐ฆฅ That's why we at PoRTaL are training our robot HAL to all the running around for you!
https://t.co/ZQRZgnz5pg
Our paper on *exponentially* faster algorithms for inverse reinforcement learning was accepted to #ICML2023! I'll put together a full explainer thread later but check out https://t.co/rid4UQdIt9 if you want a sneak peak. Makes a huge difference on hard exploration problems.
Our paper on a new (lazy) approach to model-based RL that is both computationally efficient and avoids the objective mismatch problem has been accepted for ICML!
Excited to present it at Honolulu this summer!
https://t.co/ENKppPA6Bk
Thanks everyone for coming to PORTALโs first holiday social! We had a blast showing you our wonderful robots, HAL and DORA, and our plans to take over the world ;) If you missed the demo, you can always catch it here https://t.co/7zvq9xJYbf
Happy Holidays! โ๏ธ ๐
Cleaning up after Thanksgiving dinner can be hard.. Thatโs why we are building a home robot apprentice, HAL, to clean up after you! Follow us as we teach HAL to make our lives easier so we can.. er.. get back to our phones? ๐คจ
https://t.co/10WLj4UQ6j
The PoRTaL is now open ๐
Our mission is to equip robots with skills to work seamlessly alongside human partners. We focus on imitation learning, decision making, and human-robot interaction. We are excited to get started.
Follow us along on our journey!