An Adaptive Imitation Learning Framework for Robotic Complex Contact-Rich Insertion Tasks

Complex contact-rich insertion is a ubiquitous robotic manipulation skill and usually involves nonlinear and low-clearance insertion trajectories as well as varying force requirements. A hybrid trajectory and force learning framework can be utilized to generate high-quality trajectories by imitation learning and find suitable force control policies efficiently by reinforcement learning. However, with the mentioned approach, many human demonstrations are necessary to learn several tasks even when those tasks require topologically similar trajectories. Therefore, to reduce human repetitive teaching efforts for new tasks, we present an adaptive imitation framework for robot manipulation. The main contribution of this work is the development of a framework that introduces dynamic movement primitives into a hybrid trajectory and force learning framework to learn a specific class of complex contact-rich insertion tasks based on the trajectory profile of a single task instance belonging to the task class. Through experimental evaluations, we validate that the proposed framework is sample efficient, safer, and generalizes better at learning complex contact-rich insertion tasks on both simulation environments and on real hardware.

Payload Identification and Gravity/Inertial Compensation for Six-Dimensional Force/Torque Sensor with a Fast and Robust Trajectory Design Approach

In the robot contact operation, the robot relies on the multi-dimensional force/torque sensor installed at the end to sense the external contact force. When the effective load and speed of the robot are large, the gravity/inertial force generated by it will have a non-negligible impact on the output of the force sensor, which will seriously affect the accuracy and effect of the force control. The existing identification algorithm time is often longer, which also affects the efficiency of force control operations. In this paper, a self-developed multi-dimensional force sensor with integrated gravity/inertial force sensing function is used to directly measure the resultant force. Further, a method for the rapid identification of payload based on excitation trajectory is proposed. Firstly, both a gravity compensation algorithm and an inertial force compensation algorithm are introduced. Secondly, the optimal spatial recognition pose based on the excitation trajectory was designed, and the excitation trajectory of each joint is represented by a finite Fourier series. The least square method is used to calculate the identification parameters of the load, the gravity, and inertial force. Finally, the experiment was verified on the robot. The experimental results show that the algorithm can quickly identify the payload, and it is faster and more accurate than other algorithms.

Force Control Strategy of Five-Digit Precision Grasping With Aligned and Unaligned Configurations

Objective To investigate the digit force control during a five-digit precision grasp in aligned (AG) and unaligned grasping (UG) configurations. Background The effects of various cylindrical handles for tools on power grasp performance have been previously investigated. However, there is little information on force control strategy of precision grasp to fit various grasping configurations. Method Twenty healthy young adults were recruited to perform a lift-hold-lower task. The AG and UG configurations on a cylindrical simulator with force transducers were adjusted for each individual. The applied force and moment, the force variability during holding, and force correlations between thumb and each finger were measured. Result No differences in applied force, force correlation, repeatability, and variability were found between configurations. However, the moments applied in UG were significantly larger than those in AG. Conclusion The force control during precision grasp did not change significantly across AG and UG except for the digit moment. The simulator is controlled efficiently with large moment during UG, which is thus the optimal configuration for precision grasping with a cylindrical handle. Further research should consider the effects of task type and handle design on force control, especially for individuals with hand disorders. Application To design the handle of specific tool, one should consider the appropriate configuration according to the task requirements of precision grasping to reduce the risk of accumulating extra loads on digits with a cylindrical handle.