Robot Arm Configuration Self-Awareness via Interacting with Environment

Controlling robotic arm reaching is a highly challenging research topic. Since sampling is time-consuming and cumbersome, a stream of work has resorted to utilizing forward kinematic models to provide data for training arm control models. However, existing forward models and control models are fraught with difficulties once the configuration of a robotic arm significantly changes, requiring either the re-measurement or large-scale data collection to retrain the network. Consequently, Self-awareness of the current arm configuration is of vital importance, especially after changes in the arm configurations. To swiftly obtain a high-precision forward model following arm configuration changes, we present a framework to construct the forward model based on Denavit-Hartenberg (D-H) model. Distinct from precious methods that build a forward model using a neural network, the linear relationship between the position of the end-effector and the D-H parameters of the robot arm is employed to construct the forward kinematic model. Through interacting with the environment, the robot is capable of adapting to its current state in real-time and generating a large amount of training data for control models. Extensive experimentation on the humanoid robot PKU-HR6.0 validates the efficacy of the proposed framework.