From Simulation to Hardware: Adaptive Walking of a Wireless Bipedal Robot Using Reinforcement Learning
IET Cyber-Systems and Robotics · 2026-01 · 8 (1)
DOI: 10.1049/csy2.70073 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Cyber-Systems and Robotics, online erschienen am 06.11.2026. Verfasst von Sindhu Hak Gupta und 4 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This paper presents the design and development of a wireless bipedal robot point contact (POCO). The proposed robot combines mechanical innovation with learning‐based control to improve stability during operation. The main emphasis is laid on hardware development, specifically, a magnetic suspension system that provides passive compliance, impact damping and improved balance in dynamic walking as opposed to conventionally rigid bipedal platforms. The fundamental contribution of this work is that the proposed system decreases mechanical shock at the contact point of feet of the robot and the ground, as well as enhances robustness to external disturbances. Autodesk Fusion 360 is used to design the mechanical structure, and multi‐joint dynamics with contact (MuJoCo) is used to simulate the structure and optimise the stability, joint dynamics and performance of the structure before fabrication. Furthermore, three‐dimensional (3D) printing and rapid prototyping enable refinement at the structural level through iterative frameworks. The robot uses cost‐effective high‐torque servo actuators and a four‐bar linkage mechanism together to improve the transmission of torque, create smoothness in motion, and provide mechanical efficiency without being heavyweight. The introduced magnetic suspension provides active compliance at critical joints, thus lowering the robot's dependence on software for stability and enhancing its overall inherent mechanical stability. The proposed hardware design offers a powerful platform for dynamic locomotion. Reinforcement learning (RL) is used to optimise the gait in order to supplement the mechanical design. The policies learnt through simulation are deployed to the hardware via a simulation‐to‐reality (sim‐to‐real) pipeline, which enables adaptive gait control and improved disturbance rejection in line with the principles of hybrid zero dynamics. In general, the proposed POCO robot indicates that magnetic suspension combined with learning‐based control is a successful method for making a stable and affordable bipedal platform.
Zitieren
APA
Gupta, S. H., Cheng, X., Giri, Y., Tak, O., & Sah, A. (2026). From Simulation to Hardware: Adaptive Walking of a Wireless Bipedal Robot Using Reinforcement Learning. IET Cyber-Systems and Robotics, 8(1). https://doi.org/10.1049/csy2.70073
Vancouver
Gupta SH, Cheng X, Giri Y, Tak O, Sah A. From Simulation to Hardware: Adaptive Walking of a Wireless Bipedal Robot Using Reinforcement Learning. IET Cyber-Systems and Robotics. 2026;8(1). doi:10.1049/csy2.70073
BibTeX
@article{Gupta2026,
author = {Gupta, Sindhu Hak and Cheng, Xiaochun and Giri, Yogesh and Tak, Ojasvi and Sah, Apoorva},
title = {From Simulation to Hardware: Adaptive Walking of a Wireless Bipedal Robot Using Reinforcement Learning},
journal = {IET Cyber-Systems and Robotics},
year = {2026},
volume = {8},
number = {1},
doi = {10.1049/csy2.70073}
}
EUVIMED