本文探讨了机器人自我反省(Robotintrospection)、基于LSTM的变分自回归(LSTM-VAR)等技术在机器人任务执行中的应用,包括物体识别(ObjectIntrospection)、装配任务(Kittingtask)特征抽取、物体运动预测(Objectfallingdown)及基于IProMP与EKFSLAM的物体感知运动生成。深入讨论了蔬菜切割(Chopvegtable)技能传授、人类关怀(Humancaresing)场景下的深度模仿(Deepmimicking)以及人类行为预测(Humananticipation)。
Robot introspection
- LSTM-VAR
- LSTM-beta-VAR
Object Introspection
- Kitting task
- Extract features/transformation
- ex. Object falling down
Motion generation
- IProMP+EKF SLAM
- Object-aware motion generation
- Chop vegtable
- transfer skill to others by teaching once
Human caresing
Deep RL
- Recovery
- Deep mimicking
- sample effectioncy
- better understanding
Human anticipation/prediction
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
