Amazon today Annoudéd that a team from the University of Michigan has won Alexa Prize Simbot Challenge. Simbot Challenge’s goal is to promote the development of the next generation’s virtual assistants who help people carry out tasks in the real world by learning continuously.
Teams competing in the interactive university challenge developed virtual robots in an engaging puzzle that customers could call on with a prompt.
The University of Michigan’s Seagull Team of nine students, advised by Professor Joyce Chai, earned $ 500,000 for his first performance. Their work with the other participants is now trapped in a number of research articles.
Seagull was recognized for his Simbot’s expertise in providing an engaging experience of receiving and appropriate answers to user requests and effectively performing the desired tasks to complete the missions. The judges found that Seagull’s Simbot complete tasks of relative ease, were able to understand complex commands and provided excellent guidance and suggestions.
“Winning Simbot Challenge is a willingness to our team’s unwavering dedication and endurance,” said Seagull Team leader Yichi Zhang, a Ph.D. -Students in computer science and technique. “Each team member contributed their expertise to develop different components of the system, which ensures that our bot is really functional. To see all these components seamlessly meet and act well is incredibly rewarding.”
Alexa Prize teams help as Seagull solve long long challenges in robotics, human-ai-interaction and conversation embodied AI.
Amazon provided the Simbot Challenge participants with training data, software tools, machine learning models and the device-based 3-D embodied AI simulator Alexa Arena. The teams used these inputs to innovate, launch and experiment with their new AI ideas online and improve their research throughout the competition.
In their paper, the team nudged that they set out to create an “interactive embraced agent … which can perform complex tasks in the arenasimulation environment through dialogue with users.” To achieve this, the team was dependent on “a modular system that neural and symbolic components combine”; A “natural language understanding module [that] Applying a hierarchical pipeline to convert users users to logical symbolic representations of their intentions and semantics “and” a neural vision module [that] Record object classes, states and space conditions. The “team” also developed tools and pipelines to increase our vision and language data, which constantly improves the robustness and performance of our system. “
“Alexa Prize teams, like Seagull, help to solve long-length challenges in robotics, interaction between people and AI and conversation embodied AI,” said Reza Ghanadan, a senior main researcher at Alexa AI and head of the Alexa award. “One meaning, this research is that it can potentially lead to the development of new mechanisms to create more robust A AI models that are by nature groups in the real world, drives disadvantage in the environment and can cooperate safely with man to end complex.”
“This challenge taught us that this generalization is really the key to the next generation of bodily AI,” said the Seagull team, who leads Jianing (Jed) Yang, who is also a PhD student in computer science and technique in Michigan. “One can add a lot of heuristics very quickly to achieve a perfect score on an existing task, but to be able to generalize to unseen tasks and about the real difficulties.”
Five University Teams We chose to participate in the last live interaction phase of Alexa Prize Simbot Challenge, which took place last spring. Teams from the University of California (UC), Santa Barbara and UC Santa Cruz were awarded $ 100,000 for second and $ 50,000 for third place, respective.
“To develop the next generation of embodied robotic assistants, it is important to prioritize user-centricity and proactivity in human-robot interactions,” said Jiachen Li, a first-year PhD student and team leader for UC Santa Barbara’s Gauchoai. “We learned that this means that our robots should go beyond just following human instructions and also having the ability to anticipate the user’s intention during these interactions.
The Gauchoai team was advised by Xifeng Yan, Narayanamurti Professor of Computer Science at UC Santa Barbara. Xin (Eric) Wang, assistant professor in computer science and technique, advised UC Santa Cruz’s Slugjarvis team.
“The skills and technologies developed for Simbot Challenge have applications in the real world,” said Jing GU, a first-year PhD student at UC Santa Cruz and head of Slugjarvis. “Our team’s success can lead to opportunities in home automation, robotics and AI. Participating in the challenge and reaching the final is a valuable learning experience.”