Carlsbad, CA – Army and Navy Academy is celebrating an outstanding academic milestone by Cadet Cohen Lu, whose AI research paper was accepted and presented at the NeurIPS 2025 Workshop on SPACE in Vision, Language, and Embodied AI (SpaVLE) on December 7, 2025, at the San Diego Convention Center.
Cohen participated in a summer research experience at The Berkeley Artificial Intelligence Research (BAIR) Lab, where he collaborated with his sister, a current student at the University of Illinois Urbana-Champaign, and an undergraduate mentor on a robotics-focused research project. Their work explored a practical challenge in robotics: how to train robot arms to interact safely and effectively with objects, while reducing the time, cost, and risk associated with traditional training methods.
In many robotics systems, early-stage training can require extensive human demonstration, including recording large sets of videos from multiple angles or physically guiding robotic arms through repeated movements. Beyond the time commitment, this approach can be risky because unrefined models may move unpredictably and damage hardware or nearby objects.
To address these challenges, Cohen’s team explored a simulation-first approach and introduced a VR-supported method that simplifies how robots learn object interaction and movement planning. By using a VR headset and motion tracking, the team demonstrated object identification and movement intent in a controlled virtual environment. This approach helps reduce the need for exhaustive manual demonstrations and supports safer training before real-world deployment.
“Instead of manually moving the robot arm or recording tons of videos, we showed the AI what objects were and how to move around them using our own hands and gestures inside a VR environment,” Cohen explained.
The SpaVLE workshop’s mission is to bring together historically separate research efforts across natural language processing, computer vision, and robotics, advancing research on spatial understanding and representation. Cohen’s acceptance and presentation at SpaVLE is a rare achievement for a high school student, and a meaningful example of what is possible when students are challenged with advanced, real-world work and supported by mentorship.
“Cohen’s accomplishment reflects the kind of curiosity, discipline, and academic courage we work to cultivate at Army and Navy Academy,” said Barry Shreiar, President of Army and Navy Academy. “He pursued a complex problem, contributed to real research, and represented himself and our community with maturity on a world-class stage.”
As a next step, Cohen and his team plan to continue the simulation project in the summer of 2026, aiming to develop and test a simulation system to bring VEX robots to RoboVerse in order to optimize the design and reduce the rebuild cost for VEX competitors.
For more information about NeurIPS 2025, visit:
https://space-in-vision-language-embodied-ai.github.io
For more information about The Berkeley Artificial Intelligence Research (BAIR) Lab, visit: