Configuration and Force-field Aware Variable Impedance Control with Faster Re-learning

• Published in: Journal of intelligent & robotic systems 2024-03, Vol.110 (1), p.3, Article 3
• Main Authors: Jadav, Shail, Palanthandalam-Madapusi, Harish J.
• Format: Article
• Language: English
• Subjects: Adaptation; Algorithms; Artificial Intelligence; Control; Electrical Engineering; Engineering; High impedance; Liapunov functions; Mechanical Engineering; Mechatronics; Regular Paper; Robot learning; Robotics; Stabilization; Tracking errors
• ISSN: 0921-0296; 1573-0409
• DOI: 10.1007/s10846-023-02022-x

Variable impedance control (VIC) is rapidly becoming an important ingredient for robotic manipulation in unstructured and uncertain environments. In such situations, it is often necessary to rapidly adapt to different impedance levels as per the task requirements, and to return to a low baseline impedance for safety requirements. Such a capability is crucial to stabilize interactions in divergent force fields, which commonly arise in a variety of contact and force production tasks and occasionally in non-contact tasks. Conventional methods, such as iterative learning control, often underperform in terms of stabilization and efficacy. While VIC algorithms perform better, typical challenges in such methods include unnecessarily high impedance adaptation in divergent fields, difficulty in distinguishing between error-independent and error-based divergent forces, and reliance on the Jacobian inverse which diminishes performance near singularities. In this paper, we introduce an innovative VIC algorithm that addresses typical VIC challenges. The proposed method employs a Cartesian-space field adaptation avoiding the need for inverting the Jacobian during adaptation, while at the same time providing a theoretical stabilization guarantee. Utilizing the Lyapunov function, the algorithm is shown to drive tracking errors to zero, even in the presence of divergent position and velocity-error fields and error-independent forces. Notably, the system exhibits human-like relearning at a faster pace when exposed to previously learned fields or perturbations, improving learning speeds by up to 47.97%. Performance validation was conducted through simulations on a two-link serial chain manipulator that mimics the human arm, as well as tests on a seven degrees-of-freedom KUKA robot, underscoring the algorithm’s advantages in handling VIC challenges and uncertain conditions.