Comparative Study on Agility, Efficiency, and Impact Absorption of Bipedal Robots with Active Toes

While human feet and toes are essential for efficient, agile, and stable locomotion, most bipedal robots utilize simplified foot structures to prioritize mechanical durability. This study investigates whether incorporating an active, human-inspired toe joint into a 14-DOF bipedal robot can quantitatively improve locomotion performance in terms of energy efficiency, impact absorption, and agility.

The researchers developed a 14-DOF bipedal robot featuring a "HyperLeg" mechanism, which includes a 2-DOF ankle and a 1-DOF active toe. To ensure a rigorous comparison, they created a high-fidelity simulation environment that models actual actuator dynamics, including coupled transmissions and power consumption. They trained two versions of the robot—one with active toes and one with the toes removed (toe-ablation)—using identical reinforcement learning (RL) procedures and a minimal reward function. The study specifically measured the Cost of Transport (CoT), ground reaction forces (GRF) during heel-strike, and path deviation during a custom "Robot T-Test" agility course.

The simulation results demonstrate that the active toe configuration outperforms the toe-ablation model across all key metrics. At a walking speed of 1.33 m/s, the toe-equipped robot achieved a 17.5% reduction in CoT and a 5.0% reduction in heel-strike GRF, indicating improved energy efficiency and better impact management. Furthermore, in the agility test, the active toe significantly improved path tracking, with average and maximum path deviations decreasing by 25.0% and 34.0%, respectively. These findings suggest that active toes are a viable design choice for closing the efficiency gap between robotic and human locomotion.

This study provides a quantitative foundation for the benefits of complex foot mechanics in bipedal robotics. By demonstrating that active toes contribute to both energy savings and maneuverability, the research offers a compelling argument for moving beyond simple, rigid foot designs. The methodology, particularly the use of high-fidelity actuator modeling and direct CoT minimization in RL, provides a framework for future researchers to bridge the sim-to-real gap in legged robotics.