Fysiverse-UniWAM: Unified World Action Model for Navigation and Manipulation
Overview
Fysics AI与复旦大学联合推出Fysiverse-UniWAM，整合导航、移动操作与固定工位灵巧操作，无需大规模预训练，实现24项真实任务的高效执行。模型采用MAP与Mixed-Stream架构，构建7500小时数据集，部署至4种机器人平台。","latest":"模型无需大规模预训练，通过MAP与Mixed-Stream架构实现导航与操作联合学习，完成24项真实任务测试。"}
Generated from attributed reports · Updated 20 minutes ago
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · trajectory_count: 1500000 trajectories · trajectory count
“团队从预训练视频世界模型出发，通过共享骨干网络实现导航与操作的联合学习，并提出 MAP（Manipulation Anchor Pose，操作锚定位姿），构建了超过 150 万条轨迹、累计约 7500 小时的 MAP-Data 数据集 。”
Exact source · revision 2Source owner not reported
“团队从预训练视频世界模型出发，通过共享骨干网络实现导航与操作的联合学习，并提出 MAP（Manipulation Anchor Pose，操作锚定位姿），构建了超过 150 万条轨迹、累计约 7500 小时的 MAP-Data 数据集 。”
Exact source · revision 1jiqizhixin
Reported quantity · aggregate_sensor_hours: 7500 hours · Basis not reported
“团队从预训练视频世界模型出发，通过共享骨干网络实现导航与操作的联合学习，并提出 MAP（Manipulation Anchor Pose，操作锚定位姿），构建了超过 150 万条轨迹、累计约 7500 小时的 MAP-Data 数据集 。”
Exact source · revision 2Source owner not reported
“团队从预训练视频世界模型出发，通过共享骨干网络实现导航与操作的联合学习，并提出 MAP（Manipulation Anchor Pose，操作锚定位姿），构建了超过 150 万条轨迹、累计约 7500 小时的 MAP-Data 数据集 。”
Exact source · revision 1jiqizhixin
Reported quantity · Reported success rate: 80 percent · Basis not reported
“UniWAM-Scaled 在 4 项移动操作任务中的成功率全部第一，平均成功率达到 80.0%，领先当前最强的机器人预训练基线 π₀.₅达 17.5 个百分点。”
Exact source · revision 2Source owner not reported
“UniWAM-Scaled 在 4 项移动操作任务中的成功率全部第一，平均成功率达到 80.0%，领先当前最强的机器人预训练基线 π₀.₅达 17.5 个百分点。”
Exact source · revision 1jiqizhixin
Reported quantity · average_position_error: 9.85 other · Basis not reported
“在 MAP-Bench 基准测试中，Fysiverse-UniWAM 的平均位置误差仅为 9.85 厘米 ，平均朝向误差为 2.30° ，较最强外部基线分别降低 30.1% 和 44.0% 。”
Exact source · revision 1jiqizhixin
Reported quantity · average_orientation_error: 2.3 other · Basis not reported
“在 MAP-Bench 基准测试中，Fysiverse-UniWAM 的平均位置误差仅为 9.85 厘米 ，平均朝向误差为 2.30° ，较最强外部基线分别降低 30.1% 和 44.0% 。”
Exact source · revision 1jiqizhixin
Report timeline
Follow attributed reports and material updates.
- Machine Heart — Robotics on WeChatSignalFysiverse-UniWAM: A Unified World Action Model for Navigation and Manipulation Without Large-Scale Pre-Training
Fysiverse-UniWAM, developed by Fysics AI in collaboration with Fudan University, presents a unified world action model that integrates navigation, mobile manipulation, and fixed-position dexterous tasks. The model avoids traditional large-scale robot pre-training by using MAP (Manipulation Anchor Pose) and Mixed-Stream architecture, achieving high performance on 24 real-world tasks with significant improvements over existing baselines.
- Chinese robotics — Research news discoverySignalFysiverse-UniWAM: A Unified World Action Model for Navigation and Manipulation Without Large-Scale Robot Pre-Training
A research team from Fysics AI and Fudan University's CITLab introduces Fysiverse-UniWAM, a unified world action model that integrates navigation, mobile manipulation, and fixed-position dexterous tasks. The model achieves strong performance on 24 real-world tasks, with notable improvements in positioning and orientation accuracy compared to existing baselines. The approach uses MAP (Manipulation Anchor Pose) and Mixed-Stream architecture to enable joint learning of navigation and manipulation, demonstrating the potential of alternative methods to large-scale robot pre-training.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).