Optimizing Real-World PID Control for Walking Quadruped Soft Robots via Genetic Algorithms

What is it about?

Building models and controlling soft robots has always been quite a challenge—it’s not easy to get them to move precisely. To solve this problem, we’ve designed a new walking control method for quadruped soft robots. Put simply, we use a "genetic algorithm" to optimize the traditional PID controller, making the robot’s walking more reliable. Here’s how we did it: First, we built a control model through geometric analysis, combining the robot’s unique leg structure and bend sensor. This model links "valve voltage" to "leg bending degree"—it’s like setting a clear set of "movement rules" for the robot’s legs, which simplifies the control logic. A model alone isn’t enough, though. We also used the genetic algorithm—a smart tool that automatically adjusts the PID controller’s parameters. It finds the best combination of parameters without the need for tedious manual trial and error, which boosts control performance. To test if this new method works, we ran experiments on a physical quadruped soft robot made with 3D printing. The results were impressive: Compared to the traditional Ziegler-Nichols tuning method, our optimized controller greatly improved trajectory tracking accuracy. The robot’s walking speed increased from 5 mm/s to 8 mm/s (a nearly 60% boost), its error rate dropped by 2.4064%, its "overshoot"—think of this as how much the robot’s movement "overshoots" its target—decreased by 12.55%, and its response time was cut by 0.5 seconds. What’s more, the new method performed even better than another common method called "particle swarm optimization": The error rate fell by an additional 0.4079%, overshoot decreased by 8.4%, and response time was shortened by another 1 second. Put simply, with this method, the quadruped soft robot walks faster, more accurately, more stably, and responds more quickly.

Featured Image

Why is it important?

This paper combines the "genetic algorithm" with the mature PID controller to realize automatic parameter optimization of PID — replacing the limitations of traditional manual trial and error or classical parameter tuning methods (such as Ziegler-Nichols), and being able to quickly find the global optimal parameter combination, thus fundamentally improving the adaptability and efficiency of the control strategy. It is compared not only with the classical "Ziegler-Nichols parameter tuning method" (the benchmark for traditional PID parameter tuning) but also with the mainstream intelligent optimization method "particle swarm optimization", covering both "traditional and modern" benchmarks.