- Two California students won the LeRobot Hackathon with a robot that folds clothes and sorts medicine, aiming to solve real-world household challenges.
- They used imitation and reinforcement learning with LeRobot S0-100 arms, achieving up to 85% folding accuracy and releasing their models on Hugging Face.
- The project highlights a growing focus on robotics for everyday tasks, with the team planning to expand their work into more advanced in-home solutions.
Two young roboticists from San Jose, Calif., won first place at the recent LeRobot Worldwide Robotics Hackathon for building a robot that folds clothes and sorts medicine.
Pranav Saroha, a 16-year-old high school student at Valley Christian High School, and Shaan Patel, a 19-year-old mechanical engineering major at Santa Clara University, collaborated at the 48-hour global competition organized by Hugging Face.
They won first place in the U.S. and second worldwide, in a competition with more than 250 teams and 3,000 hackers across 44 countries.
Why did they choose everyday tasks for the robot to tackle?
“I really wanted to help out my grandfather who struggles with managing small pills,” said Saroha, in an interview with The AI Innovator. “And I don’t really like doing laundry, so I just thought it’d be pretty cool to have (a robot) just folding some T-shirts.”
The team used LeRobot S0-100 robotic arms and fine-tuned AI models to train the robot through imitation learning and reinforcement learning. The models were trained on π0 (pi zero), SmolVLA, and ACT.
The team recorded 50 training episodes for folding T-shirts using a bimanual SO-100 robotic arm and four-camera setup. Their folding demo achieved a 70% success rate during the hackathon, which Saroha later improved to 85% using a different model.
They also trained the robot to sort pills into daily medicine boxes – an application they noted could help elderly users.
The team’s training data and models have since been released on Hugging Face. The training setup involved synchronized camera feeds and motor position data, feeding into vision-language-action models to fine-tune behavior.
But the team ran into some hiccups – such as an unexpected API upgrade that caused issues when they tried to push the dataset to Hugging Face via automated scripts. “It caused a lot of problems,” Saroha said.
The team had to manually upload training data and reverse recent changes to Hugging Face repositories during the night.
They also lost 16 episodes due to a camera cable disconnection. “Always check your hardware again and again,” Saroha said.
Despite the setbacks, Patel said he took away some valuable lessons from the hackathon. “They pretty much taught me all the different things that you can come across,” he said. “This whole thing was a learning experience, and I’m super grateful for it.”
Organizers said the hackathon wanted to shine a spotlight on robotics.
“One of the reasons this hackathon was organized was because we want to call the ‘ChatGPT’ moment for robotics,” said Dhruv Diddi, founder and CEO of Solo Tech, which hosted the Bay Area node of the hackathon. “Many people think it’s not going to come from some fancy industrial use of robotics but something that helps you around at home.”
Priya Shivakumar, chief product officer of Lightning AI, which provided the contest’s compute resources, agreed. “What’s most exciting is they focused on the everyday tasks, like folding T-shirts and sorting medicine,” she said. But “it’s not just their vision that stood out, but it was really the execution. They built, trained and deployed everything in just 48 hours.”
Saroha plans to keep building on the project. “I already have some idea of how to create a seven-degree-of-freedom arm,” he said, referring to the number of joints. He is also working on generating synthetic episodes to expand training without having to manually controlling the robot.
Patel, who is interning at Solo Tech, is working on a few things, including figuring out a new way to use robots in manufacturing.
Both Saroha and Patel plan to continue developing home-use robotic solutions. “This is just the beginning,” Saroha said.
