Unitree Astra + Thinker-VLA Embodied Intelligence Experiments
Overview
Ubtech (HK9880) conducted six experiments combining the general-purpose Astra model with its proprietary Thinker-VLA and Thinker-WM models to explore embodied intelligence. Real-world tests showed that while Astra can understand scenes and plan tasks, direct control often fails at fine motor skills due to high computational costs. Integrating Thinker-VLA for execution and Thinker-WM for future prediction improved efficiency and task success rates, illustrating a path toward scalable robotic applications.
From Chinese robotics — Robot learning research
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · model processing time: 70 other · Basis not reported
“本次实验记录： 模型累计处理耗时：约 70 分钟 Token 消耗：约 9557.8 万 Astra API 估算费用：约 579 美元 最终结果：未完成抓取 这揭示了通用大模型直接控制机器人的现实挑战：它能够理解空间关系、生成动作决策，但面对连续运动和精细操作时，仍可能需要大量推理与反复修正。”
Exact source · revision 2Source owner not reported
Reported quantity · Astra API estimated cost: 579 other · Basis not reported
“本次实验记录： 模型累计处理耗时：约 70 分钟 Token 消耗：约 9557.8 万 Astra API 估算费用：约 579 美元 最终结果：未完成抓取 这揭示了通用大模型直接控制机器人的现实挑战：它能够理解空间关系、生成动作决策，但面对连续运动和精细操作时，仍可能需要大量推理与反复修正。”
Exact source · revision 2Source owner not reported
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Robot learning researchUbtech Demonstrates Embodied AI Synergy with Astra Model and Proprietary Thinker Models
Ubtech (HK9880) conducted six experiments combining the general-purpose Astra model with its proprietary Thinker-VLA and Thinker-WM models to explore embodied intelligence. Real-world tests showed that while Astra can understand scenes and plan tasks, direct control often fails at fine motor skills due to high computational costs. Integrating Thinker-VLA for execution and Thinker-WM for future prediction improved efficiency and task success rates, illustrating a path toward scalable robotic applications.
Event coverage history
Current attention 4·Peak within the comparable range 5（2026-10-31 02:00 UTC）·Change within the comparable range over 24 hours –
The trend compares participants observed continuously; its scope may be narrower than current attention totals. Hover or click to explore hourly observations, or use the left and right arrow keys.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).