Most quadruped robots rely on dedicated controllers to walk, rather than exploiting the mechanical resonance that animals use to move efficiently. A team from DLR and other institutions proposes an alternative: make the robot's nonlinear dynamics predictable during design, then let gaits emerge from those dynamics. Their proof-of-concept robot, eBert, is highly compliant, with elasticity and inertia shaping its motion.
The researchers identified six nonlinear normal modes (NNMs) on eBert using new theoretical tools, then validated these modes in both simulation and hardware. By applying black-box optimization to adjust step length, each NNM naturally developed into a distinct gait with its own speed. The resulting movements largely transferred from simulation to the physical robot, showing that the mechanics themselves can generate task-specific locomotion.
The work suggests that embodied intelligence—letting body morphology and passive dynamics do the work—could be a foundation for a new generation of agile, efficient quadruped robots. Since the paper is a single source, there are no conflicting findings to compare; the authors frame the results as a proof of concept rather than a finished system.