Author(s)

ShuoPei Yang^*, ChangCheng Zhao

Affiliation(s)

Jiangsu Xingzhitu Intelligent Technology Co., Ltd., Suzhou 215000, Jiangsu, China.

Corresponding Author

ShuoPei Yang

ABSTRACT

This study addresses the challenge of adaptive gait planning for quadrupedal robots in complex terrains by proposing a reinforcement learning-based solution. First, the kinematic model of the quadruped robot and the complex terrain model are established, providing a theoretical foundation for subsequent algorithm design. Second, a hierarchical reinforcement learning framework is introduced, comprising a high-level gait policy and a low-level joint control policy, to accommodate varying locomotion demands across different terrains. Additionally, an adaptive exploration mechanism and a safety layer based on control barrier functions are incorporated to ensure efficient exploration and operational safety.The proposed algorithm demonstrates robust gait performance across diverse terrains, exhibiting notable advantages in motion performance, adaptability, and computational efficiency. Specifically, simulation results highlight improvements in terrain adaptability and gait stability, while hardware experiments further validate the feasibility and effectiveness of the method in real-world applications.Compared to existing approaches, the main innovations of this study lie in the incorporation of a curriculum learning-based strategy for progressively increasing terrain difficulty and an uncertainty-driven exploration reward mechanism. These designs significantly enhance the adaptive capability of the robot in complex environments. However, the algorithm still faces limitations in computational complexity and real-time performance. Future research may focus on optimizing the algorithmic structure to achieve more efficient real-time control.In summary, this work offers an effective solution for adaptive gait planning of quadruped robots in complex terrains, with both theoretical significance and practical value.

KEYWORDS

Quadruped robots; Reinforcement learning; Complex terrains; Motion planning

CITE THIS PAPER

ShuoPei Yang, ChangCheng Zhao. Adaptive gait planning for quadruped robots in complex terrains via reinforcement learning. World Journal of Information Technology. 2025, 3(3): 78-88. DOI: https://doi.org/10.61784/wjit3056.

REFERENCES

[1] Guo J, Yu J, Feng C, et al. Global path planning for robots on uneven terrain based on improved A* algorithm. Computer Engineering and Applications, 2025, 61(5): 309-322.

[2] Liu S, Zhang Z, Li Z, et al. Autonomous path planning for companion robots based on bidirectional regional RRT*. Computer Engineering and Applications, 2025, 61(4): 1-12.

[3] Luo Z, Han Y, Zhang X, et al. Research on path planning of inspection robot based on improved RRT-Connect and DWA algorithm. Journal of Computer Engineering and Applications, 2024, 60(15).

[4] Chen K, Tian B, Li H, et al. Research on motion control of two-wheeled legged robot based on DDPG algorithm. Systems Engineering and Electronics, 2023, 45(4): 1144-1151.

[5] Dong H, Yang J, Li S, et al. Research progress in robot motion control based on deep reinforcement learning. Control and Decision, 2022, 37(2): 278-292.

[6] Nahrendra IMA, Yu B, Myung H. Dream WaQ: Learning robust quadrupedal locomotion with implicit terrain imagination via deep reinforcement learning. Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2023: 5078-5084.

[7] Xu W, Jin X, Zhang L, et al. Research on autonomous navigation of coal mine mobile robot based on multi-sensor data fusion. Coal Mine Machinery and Electrical, 2023, 44(1): 8-12.

[8] Kang D, De Vincenti F, Adami N, et al. Animal motions on legged robots using nonlinear model predictive control. Proceedings of the International Conference on Intelligent Robots and Systems (IROS), 2022: 11955-11962.

[9] Wensing PM, Posa M, Hu Y, et al. Optimization-based control for dynamic legged robots. IEEE Transactions on Robotics, 2023, 40: 43-63.

[10] Chen P, Pei J, Lu W, et al. A deep reinforcement learning based method for real-time path planning and dynamic obstacle avoidance. Neurocomputing, 2022, 497: 64-75.

[11] Avila-Campos P, Haxhibeqiri J, Girmay M, et al. Residual service time optimization for legacy wireless-TSN end nodes. Proceedings of the 19th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), 2023: 466-471.

[12] Zhou F, Zhao J, Long H, et al. Path planning for quadruped robots based on improved RRT algorithm. Journal of Hunan University of Technology, 2024, 38(6): 10-20.

[13] Zhao Z, Tao Q, Yang T, et al. Trajectory tracking control of lower limb rehabilitation robot based on DDPG. Machine Tool & Hydraulics, 2023, 51(11): 13-19.

[14] Chen T, Li Y, Rong X. Design and verification of a real-time foot force optimization method for quadruped robots with dynamic gaits. Robot, 2019, 41(3): 307-316.

[15] Zhang P, Liao Q, Wei S, et al. Research on the control system of a hydraulically driven quadruped robot. Hydraulics & Pneumatics, 2011(1): 29-31.

[16] Wang H, Lu H, Chen L, et al. UAV multi-region path planning fusion algorithm combined with B-spline optimization. Computer Measurement & Control, 2022, 30(9): 193-200.