人机协作中人的动作终点预测

为实现安全高效的人机协作（HRC），需要机器人及时对人的动作做出预测，从而积极主动地辅助人工作。为解决在HRC装配场景中机器人对人的动作终点预测问题，提出了一种基于长短时记忆（LSTM）网络的动作终点预测方法。在训练阶段，用人的动作序列与对应的动作终点组成的样本训练LSTM网络，构建动作序列与动作终点之间的映射。在应用阶段，根据人的动作的初始部分对动作终点提前做出预测。通过在装配场景中，对人抓取工具或零件的动作终点进行预测，验证了所提方法的有效性。在观测到50%的动作片段时，预测准确率达到80%以上。 

Human motion end point prediction in human-robot collaboration

To realize a safe and effective human-robot collaboration (HRC), it is necessary for the robot to predict human motions in a timely manner, so as to assist human more actively in the cooperative work. In order to solve the problem of human motion prediction in HRC assembly scenario, a motion end point prediction method based on long short-term memory (LSTM) network is proposed. In the training phase, the LSTM network is trained with samples of human motion sequences and corresponding motion end points, and the mapping between motion sequences and motion end points is constructed. In the application phase, the motion end point is predicted in advance based on the initial part of the human motion sequence. The effectiveness of the proposed method is verified by predicting the end points of motion of a human grasping tool or part in an assembly scenario. When 50% of the motion fragments are observed, the accuracy rate of prediction is above 80%.