Openpi Comet: Competition Solution For 2025 BEHAVIOR Challenge
Junjie Bai, Yu-Wei Chao, Qizhi Chen, Jinwei Gu, Moo Jin Kim, Zhaoshuo Li, Xuan Li, Tsung-Yi Lin, Ming-Yu Liu, Nic Ma, Kaichun Mo, Delin Qu, Shangkun Sun, Hongchi Xia, Fangyin Wei, Xiaohui Zeng
2025-12-16
Summary
This paper details a team's attempt to create a robot that can perform everyday household tasks, specifically as part of the 2025 BEHAVIOR Challenge.
What's the problem?
Currently, robots struggle with tasks that require planning and executing a series of actions over a long period of time, especially in realistic home environments. Existing research often focuses on simpler tasks and doesn't translate well to the complexity of real-world situations where people actually need robotic assistance. The challenge is to build a robot that can reliably complete these longer, more complex tasks.
What's the solution?
The team started with an existing AI model called π_{0.5} and improved it by carefully experimenting with different training methods and the amount of data used. They systematically tested what worked best, focusing on both initial 'pre-training' and later 'post-training' adjustments. This allowed them to build a robot that performed very well in the BEHAVIOR Challenge, coming in second place and significantly better than other entries.
Why it matters?
This work is important because it shows how to take powerful AI models and adapt them to control robots in complex, real-world scenarios. The team shared their findings and recommendations, hoping to help other researchers build better robots that can actually be useful to people in their homes.
Abstract
The 2025 BEHAVIOR Challenge is designed to rigorously track progress toward solving long-horizon tasks by physical agents in simulated environments. BEHAVIOR-1K focuses on everyday household tasks that people most want robots to assist with and these tasks introduce long-horizon mobile manipulation challenges in realistic settings, bridging the gap between current research and real-world, human-centric applications. This report presents our solution to the 2025 BEHAVIOR Challenge in a very close 2nd place and substantially outperforms the rest of the submissions. Building on π_{0.5}, we focus on systematically building our solution by studying the effects of training techniques and data. Through careful ablations, we show the scaling power in pre-training and post-training phases for competitive performance. We summarize our practical lessons and design recommendations that we hope will provide actionable insights for the broader embodied AI community when adapting powerful foundation models to complex embodied scenarios.