Robot design adds components. More sensors, more actuators, more degrees of freedom. The assumption runs deep: a robot that can do more has more. Limbs are capabilities. Removing one is damage.
Bhatia, Karamcheti, Jayaraman, and Coros (arXiv:2603.12505, March 2026) build a robot from randomly assembled modular components, then improve its locomotion by deliberately breaking off its own limbs. Kinematic self-destruction — amputation as optimization. The robot starts with a body it didn't choose and discovers, through simulation and physical testing, that removing specific limbs makes it faster, more stable, or both.
The key result: self-destructed robots outperform not only their intact originals but also randomly pruned configurations. The improvement is not explained by simplification alone — random removal hurts. The improvement requires knowing which piece to lose. The robot finds morphological dead weight that a designer wouldn't have added but that a random assembly inevitably includes.
The through-line is that subtraction is a design strategy with its own optimization landscape, distinct from the addition landscape. Adding a limb and removing a limb are not inverse operations in functional space. A body optimized by addition and a body optimized by subtraction from a larger initial structure will generally not converge to the same morphology. The two landscapes have different basins, different gradients, different optima.
This matters beyond robotics. Evolutionary biology has long recognized vestigial structures — the appendix, the coccyx, the recurrent laryngeal nerve taking a detour around the aortic arch. These are remnants of addition. The organism cannot easily subtract because the structures are deeply integrated. The robot, built from modular components, can subtract cleanly. Modularity is the precondition for profitable amputation. And modularity is a design choice, not a given — meaning the ability to improve by subtraction must be designed into the system before the system knows what it will need to remove.