团队公开 2026 BEHAVIOR Challenge 方案：单一 VLA 模型覆盖 100 个家务任务
One Policy for 100 Household Tasks: Our Solution for the 2026 BEHAVIOR Challenge
家里机器人学做家务现在有了更清晰的训练路径：一个 checkpoint 管住 100 个任务，还能把 14 个难任务外包给 GPT-6 Astra 拿到 0.639 的 Q 值。做具身智能和 VLA 训练的工程师可以重点看它的任务迁移和降噪训练细节，比头部的纯 RL 方案更贴近实际算力。
参赛团队在 arXiv 发布 2026 BEHAVIOR Challenge 的技术方案，在 OmniGibson 仿真器的 100 个长时序家务任务上最终排名第 9，是前十名中唯一独立参赛队。方案核心是一个基于 $π_{0.5}$ 的视觉-语言-动作（VLA）模型，全部 100 个任务共用一个 checkpoint；在与 2025 年重叠的 50 个任务上 Q 值从 0.260 提升到 0.371，全部 100 个任务平均 Q 为 0.392。
技术改进集中在三方面：把任务集从 50 个迁移到 100 个，利用演示数据中的技能标注构建语义阶段，新任务的 stage embedding 从相似旧任务 warm-start；采用方差缩减的多采样 flow matching，并用噪声-算力模型自动决定每个观测的 action-expert 采样数；训练扩展到 200k 步，约等于对 211M 帧演示数据完整过一遍。评测侧搭建了云端基础设施，单 GPU 上并行运行六个仿真器且快于实时。
在学得策略之上，14 个任务被路由到 GPT-6 Astra 编码智能体，它通过观测和运动工具控制机器人，每集从一个冻结的练习会话副本启动，这 14 个任务上 Astra 的平均 Q 达 0.639，而学得策略仅 0.108。最终公开提交因只包含部分 Astra 结果，得分为 0.446 Q、27.3% 全任务成功率。
One Policy for 100 Household Tasks: Our Solution for the 2026 BEHAVIOR Challenge
We describe our solution for the 2026 BEHAVIOR Challenge: 100 long-horizon household tasks in the OmniGibson simulator. We placed 9th on the final leaderboard, the only independent team in the top 10. The core of the solution is a single vision-language-action (VLA) model, one checkpoint for all 100 tasks, built on our 1st-place 2025 solution, which is based on $π_{0.5}$. On the 50 tasks shared with 2025, its Q rises from 0.260 to 0.371; across all 100 tasks it scores 0.392. Our main conclusion from 2025 was that the model was undertrained, so most of the 2026 work went into training the model for longer, optimizing some training parameters, making each training step slightly faster and reducing the gradient noise: a migration from 50 to 100 tasks (semantic stages built from the demonstrations' skill annotations, task and stage embeddings of new tasks warm-started from similar old ones, and a 21-dimensional action space matching the teleoperation controls); variance-reduced multi-sample flow matching with a simple noise-vs-compute model that picks the number of action-expert samples per observation without ablations; and a 200k-step training run, about one pass over the 211M frames of the demonstrations. A cloud evaluation infrastructure keeps a shared record with provenance, episode diagnostics and restorable simulator states, and runs six simulators on one GPU together faster than real time. On top of the learned policy, 14 tasks are routed to a GPT-6 Astra coding agent that controls the robot through observation and motion tools, starting every episode from a frozen copy of its own practice session on training instances. Across these 14 tasks, Astra's mean Q is 0.639, against 0.108 for the learned policy. The public submission included only part of the Astra results and scored 0.446 Q with 27.3% full-task success.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include VLA, BEHAVIOR Challenge, OmniGibson, flow matching.
Source attribution: curated from arXiv cs.AI (@Ilia Larchenko, Gleb Zarin, Atindra Nair, Keerti Manney, Vitaly Kleban, Mirko Kiefer, Anjana Yodaiken). Cross-source corroboration: 8 independent sources have covered this development (including Fei-Fei Li, rohanpaul_ai, 宝玉), confirming industry-wide signal strength.
