Adaptive control solves uncertain joint stiffness in flexible robots

Model-based control of flexible joint robots traditionally requires accurate knowledge of joint compliance (stiffness). However, in real-world applications, stiffness properties change due to operating conditions, wear, and aging. A team of researchers (Kirner et al. from multiple institutions) has proposed an adaptive control approach that handles this uncertainty. Instead of relying on precise stiffness models, the method continuously updates an estimate of the nonlinear torque-deflection relationship for each joint. This is achieved through an implicit control law and a control-input-dependent regressor matrix—a departure from classical adaptive control used for rigid-joint robots. The approach is robust to errors from the motor position controller, making it practical for real hardware. The paper is accepted at IEEE/ASME AIM 2026 and is available on arXiv.

Experimental validation on a flexible joint with nonlinear stiffness characteristics confirmed that the adaptive controller effectively compensates for unknown and time-varying stiffness, maintaining performance without manual recalibration. This work has significant implications for the next generation of collaborative robots and advanced mechatronic systems where joint flexibility is common. By adapting to real-time changes, the method reduces the need for frequent model updates and improves long-term reliability. For engineers building resilient robotic systems, this approach offers a practical way to handle degradation over time, potentially lowering maintenance costs and extending operational life.