Beyond human heuristic in robotics leg conception

Humanoid robots are seen as a promising tool in providing better assistance for people, whithout changing their environment. To achieve this, obtaining a better robot capable of dynamical movement and good locomotion is critical. Most recent bipedal architectures use parallel mechanism transmission to enhance the dynamic capability of their design for the ankle (like inside the Unitree G1 or Tesla's new Optimus robot), or the knee (like inside Artemis or Digit) where bar transmission limits the foot inertia is often used. Novel hip designs
emerge, with the new atlas robot of Boston Dynamics, where the hip axes' orientation differs from that of older robot designs such as the Kawada's HRP-2. It raises some questions about the importance of the
geometries of the mechanisms inside the leg and the importance of the different conception parameters. This talk presents criteria to evaluate the novel generation of legs, a genetic algorithm approach to help
optimize the leg, and a novel bipedal robot featuring a closed-loop kinematic structure designed to enhance dynamic capabilities.