Hybrid Imitation Learning Framework for Robotic Manipulation Tasks

This study proposes a novel hybrid imitation learning (HIL) framework in which behavior cloning (BC) and state cloning (SC) methods are combined in a mutually complementary manner to enhance the efficiency of robotic manipulation task learning. The proposed HIL framework efficiently combines BC and SC losses using an adaptive loss mixing method. It uses pretrained dynamics networks to enhance SC efficiency and performs stochastic state recovery to ensure stable learning of policy networks by transforming the learner’s task state into a demo state on the demo task trajectory during SC. The training efficiency and policy flexibility of the proposed HIL framework are demonstrated in a series of experiments conducted to perform major robotic manipulation tasks (pick-up, pick-and-place, and stack tasks). In the experiments, the HIL framework showed about a 2.6 times higher performance improvement than the pure BC and about a four times faster training time than the pure SC imitation learning method. In addition, the HIL framework also showed about a 1.6 times higher performance improvement and about a 2.2 times faster training time than the other hybrid learning method combining BC and reinforcement learning (BC + RL) in the experiments.

Real-Time Whole-Body Imitation by Humanoid Robots and Task-Oriented Teleoperation Using an Analytical Mapping Method and Quantitative Evaluation

Due to the limitations on the capabilities of current robots regarding task learning and performance, imitation is an efficient social learning approach that endows a robot with the ability to transmit and reproduce human postures, actions, behaviors, etc., as a human does. Stable whole-body imitation and task-oriented teleoperation via imitation are challenging issues. In this paper, a novel comprehensive and unrestricted real-time whole-body imitation system for humanoid robots is designed and developed. To map human motions to a robot, an analytical method called geometrical analysis based on link vectors and virtual joints (GA-LVVJ) is proposed. In addition, a real-time locomotion method is employed to realize a natural mode of operation. To achieve safe mode switching, a filter strategy is proposed. Then, two quantitative vector-set-based methods of similarity evaluation focusing on the whole body and local links, called the Whole-Body-Focused (WBF) method and the Local-Link-Focused (LLF) method, respectively, are proposed and compared. Two experiments conducted to verify the effectiveness of the proposed methods and system are reported. Specifically, the first experiment validates the good stability and similarity features of our system, and the second experiment verifies the effectiveness with which complicated tasks can be executed. At last, an imitation learning mechanism in which the joint angles of demonstrators are mapped by GA-LVVJ is presented and developed to extend the proposed system.