Indicators reveal early signs of instability in autonomous robots
Indicators of resilience for autonomous control systems
Robotics
Summary
It can be hard to tell when autonomous machines like drones are about to fail because their problems start small and hidden. The authors show that special signals called resilience indicators can detect early warning signs before a robot becomes unstable. They tested these signals by damaging parts of a drone and found that the system's health could be tracked before serious issues appeared. This helps keep autonomous machines safer by spotting trouble early.
What this means in practice
- •For drone operators: Detect early signs of mechanical wear or damage in drones to prevent crashes before visible faults occur.
- •For industrial automation engineers: Monitor autonomous control systems in factories for subtle instabilities that indicate potential failures in robotic equipment.
Authors
Jasper van Beers, Da-Hwi Kim, Prashant Solanki, Coen de Visser
Abstract
As modern societies rely more on autonomous systems to facilitate daily life, assuring their safe operation is paramount. Naturally, there are many techniques available to predict and prevent system failures. However, the safety afforded by such schemes may become misaligned with the true system, which can change in unexpected ways - from partial faults to natural wear-and-tear - that subtly degrade its stability. The implications that such subtle changes have on autonomous system stability can be observed through generic indicators of resilience derived from critical slowing down, popular for anticipating catastrophic tipping points in natural systems. Here, we show how one can systematically design these generic indicators for nonlinear control systems and show how these can reflect loss of stability though simulations of canonical robotic systems wherein their proximity to instability is manipulated directly. These results are affirmed through real-world flight experiments of a quadrotor that is nudged towards instability by progressively damaging its propeller blades. Our results show that the implications of degraded resilience on closed-loop stability are evident well before they appear, for which the indicators of resilience derived here can provide an early warning.