US researchers have developed a new bipedal robot design, the HybridLeg platform, to advance reinforcement learning.
Equipped with a lantern-shaped sensored mechanical cover, these robots can safely handle full-body contact.
To address the inherent instability of humanoids, the University of Illinois’ Kinetic Intelligent Machines Lab (KIMLAB) has developed a protective design that cushions the impact of a fall, allows for autonomous recovery to a standing position, and can self-reset with each attempt.
This platform, combined with multimodal fall detection and enhanced stance phase tracking, paves the way for robust, long-term, real-world reinforcement learning experiments.
Rethinking bipedal design
A video shared by KIMLAB details an innovative untethered bipedal robot with a unique “hybrid leg” mechanism.
While traditional humanoid robots typically use series links to mimic human anatomy, this design combines the biological familiarity of series links with the mechanical advantages of parallel links, such as high speed, low inertia, and excellent payload-to-weight ratio.
The hybrid leg is a parallel linkage where each link consists of a series chain, forming a five-bar closed link. The robot relies on 12 motors to operate, and an important design choice is to focus 10 of them near the pelvis, leaving just two at the ankles. This configuration significantly reduces distal mass, minimizes the negative effects of swing leg dynamics, and allows for more accurate physical modeling using lower order models such as linear inverted pendulums.
The robot is a completely self-contained, disconnected platform that houses all necessary components inside its body, including a single-board computer, IMU, voltage converter, and LiPo battery. The presentation concludes by demonstrating the robot’s capabilities through various walking experiments.
Agile hybrid bipedalism
This large bipedal robot demonstrates how hybrid mechanical design can push humanoid locomotion toward greater agility, strength, and efficiency. The robot, detailed in a recent paper, is built around the HybridLeg mechanism, a novel approach that combines series and parallel structures to achieve six degrees of freedom per leg while maintaining low inertia and a large working space.
This design allows for faster motion, higher payload, and improved dynamic performance, which are key requirements for agile bipedalism. To further increase the rigidity and precision of the structure, the latest version of the HybridLeg is manufactured using carbon fiber tubing and precision bearings, allowing the structure to support its own weight without sacrificing precision.
A pair of HybridLeg is assembled into a complete bipedal platform using a custom pelvic design inspired by human biomechanics. The pelvis incorporates a yaw angle offset similar to the toe-out angle of the human foot, increasing the reachable working space of the foot and improving overall stability. Simulation results detailing workspace and speed ranges are verified through hardware experiments, confirming close agreement between theory and practice.
At 1.84 meters (1.84 meters) tall, the robot is taller than the average human and weighs just 29 kilograms (64 pounds). Despite its size, it can be driven by the same class of servo motors typically used in small humanoid robots, highlighting the efficiency and optimized structural design of the hybrid mechanism. This paper provides a detailed description of the mechanical architecture along with a complete kinematic analysis and analytical solution.
Performance validation includes preliminary hardware experiments such as crouching and walking in place, as well as multibody dynamics simulations. The researchers say a demonstration of simple forward walking further confirmed the feasibility of this approach. Taken together, these results position the hybrid leg-based biped as a promising platform for future research in humanoid locomotion, scalable robot design, and real-world dynamic locomotion experiments.
