Value as Memory (VaM) for Robotic Task Progress Memory
Overview
This research introduces VaM, a reinforcement learning framework that uses recent value predictions as task progress memory to enhance robot autonomy. VaM enables robots to continuously adapt to new states during deployment by integrating offline and online learning. The method shows significant improvements in task success rates, with over 95% success in a 4-hour continuous box assembly task. The paper also highlights the effectiveness of value caching in distinguishing visually similar states and triggering corrective actions.
From Machine Heart — Robotics on WeChat
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 · Reported success rate: 95 percent · Basis not reported · Differing source assertions
“机器人终于有了「任务进度记忆」！连续干活4小时，成功率超95%”
Exact source · revision 1jiqizhixin
Reported quantity · Reported success rate: 95 percent · Basis not reported · Differing source assertions
“在纸盒装配任务中，VaM 连续自主运行 4 小时，任务成功率达到 95% 以上，展示了长时间运行中的稳定表现。”
Exact source · revision 1jiqizhixin
Reported quantity · duration: 4 hours · Basis not reported · Differing source assertions
“机器人终于有了「任务进度记忆」！连续干活4小时，成功率超95%”
Exact source · revision 1jiqizhixin
Reported quantity · duration: 4 hours · Basis not reported · Differing source assertions
“在纸盒装配任务中，VaM 连续自主运行 4 小时，任务成功率达到 95% 以上，展示了长时间运行中的稳定表现。”
Exact source · revision 1jiqizhixin
Report timeline
Follow attributed reports and material updates.
- Machine Heart — Robotics on WeChatSignalValue as Memory: Online Value-Guided Post-training for General Robot Policies
This research introduces VaM, a reinforcement learning framework that uses recent value predictions as task progress memory to enhance robot autonomy. VaM enables robots to continuously adapt to new states during deployment by integrating offline and online learning. The method shows significant improvements in task success rates, with over 95% success in a 4-hour continuous box assembly task. The paper also highlights the effectiveness of value caching in distinguishing visually similar states and triggering corrective actions.
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).