University of Michigan Wins First Alexa Prize SimBot Challenge
The University of Michigan has won Amazon’s inaugural Alexa Prize SimBot Challenge for their Seagull virtual robot project. Seagull beat out other digital robotic assistants from other educational institutions from around the world for its performance and ability to learn how to help people in the real world.
Nine University of Michigan students and their faculty advisor built Seagull, earning $500,000 from the Alexa Prize judges for winning the competition. The judges highlighted the intuitive and pleasing control design for the virtual robot. Seagull also scored well for completing tasks even with complex commands and follow-up suggestions. That fits with the team’s goal of building an “interactive embodied agent” as described in a research paper accompanying their submission.
“Winning the SimBot Challenge is a testament to our team’s unwavering dedication and perseverance,” Seagull team leader Yichi Zhang said. “Each team member contributed their expertise to develop different components of the system, ensuring our bot is truly functional. Seeing all these components seamlessly come together and perform well is incredibly rewarding.”
Amazon created the SimBot Challenge to encourage research in linking digital robots with AI. The contestants began working in early 2022 on machine-learning models that could engage with humans using natural language understanding. Each team had to submit an had to be capable of learning from human and sensory feedback to navigate virtual surroundings, handle objects, and complete tasks They then competed in the Alexa Prize’s first public benchmark phase, where their performances were ranked, including against a bigger pool of teams Amazon deemed eligible. The benchmark challenge was planned in tandem with the public release of the TEACh dataset, a collection of more than 3,000 conversations between simulated users and robots. These conversations involve discussing methods for managing household tasks, with the human providing step-by-step instructions to the robot along with simulated visual data.
“Alexa Prize teams, like Seagull , are helping to solve long-lasting challenges in robotics, human-AI interaction, and conversational embodied AI,” Alexa AI Prize head and Alexa AI senior principal scientist Reza Ghanadan. “One significance of this research is that it may potentially lead to the development of new mechanisms to create more robust AI models that are inherently grounded in the real world, operate reliably in the environment, and can collaborate safely with humans to complete complex tasks.”
Amazon narrowed the competition from ten to five groups ahead of the final competition last month. The University of California, Santa Barbara picked up $100,000 and second place for their GauchoAI, while the University of California, Santa Cruz scored third and 50,000 for their SlugJARVIS project (the UC Santa Cruz mascot is a banana slug).