The LOCOMORPH project was organized into a series of interconnected work packages, each tackling a different aspect of the relationship between body morphology and locomotion. Below is an overview of the main research activities carried out over the project's four-year duration.
The project began with a systematic study of locomotion strategies in animals. The BioMech Student Lab led this effort, analyzing the biomechanics of walking, running, and climbing in insects, lizards, and small mammals. Using high-speed motion capture and force plate measurements, the team gathered quantitative data on how body compliance and limb geometry contribute to stable movement across different surfaces. These biological findings served as design guidelines for the robot prototypes developed in later work packages.
EPFL's Biorobotics Laboratory led the theoretical work, developing formal measures of morphological computation — mathematical frameworks for quantifying how much of a locomotion task is handled by the body versus the controller. This included information-theoretic approaches that treat the body as an information-processing system, measuring the flow of signals between the environment, the body, and the brain. These measures allowed the team to compare different body designs on a principled, quantitative basis rather than relying on intuition.
The NextStep Robotics Collective developed the LocoKit modular robotics platform, which allowed rapid construction and reconfiguration of legged robots with different body shapes, limb lengths, joint arrangements, and material properties. Multiple robot configurations were built and tested, ranging from quadruped walkers to hexapod climbers. EPFL contributed bio-inspired actuator designs including compliant tendon-driven joints and variable stiffness mechanisms.
The WalkBot Initiative focused on reinforcement learning algorithms that could exploit the physical properties of the robot's body rather than fighting against them. The team developed learning methods that co-optimize the controller and the body morphology simultaneously, demonstrating that joint optimization produces significantly better locomotion performance than optimizing either one in isolation. Simulation experiments showed that robots with well-designed morphologies could learn effective gaits with far fewer training iterations.
High-fidelity physics simulation was central to the project. Custom simulation environments were developed that could accurately model soft body dynamics, contact mechanics, and material deformation. These simulators allowed the team to evaluate thousands of morphological variations that would have been impractical to build physically, identifying promising designs for hardware validation.
The final phase brought together the biological insights, theoretical frameworks, learning algorithms, and hardware platforms into integrated demonstrations. Robots with morphologically optimized bodies were shown to traverse terrain that defeated conventionally designed machines, including loose gravel, inclined surfaces, and obstacles larger than the robot's leg length. Results were shared at community meetups, maker faires, and online through open-source channels.
Throughout the project, the LOCOMORPH team published articles and tutorials, organized workshops on morphological computation, and released the LocoKit platform as open-source hardware. Public outreach activities included robotics demonstrations at maker spaces, community events, and schools.