Original title: MulDP: Multimodal Diffusion Policy for Autonomous Quadruped Parkour Navigation across Complex Terrains
Kangmai Hu, Yueqi Zhang, Peng Zhai, Xiaoyi Wei +4 · September 2026
Four-legged robots can perform impressive parkour over tough obstacles, but people still usually guide their overall path. True autonomous parkour navigation remains underexplored. Robots struggle to adjust their speeds precisely, plan far ahead, and match what they see with how their bodies move. To solve this, researchers built a system called MulDP. It combines visual inputs, the robot's sense of its own body position, and its goal to calculate smooth, forward-looking velocity commands. To train this system, the team created the first Quadruped Parkour Navigation Dataset, which features varied navigation behaviors across challenging environments. In both simulations and real-world tests, this approach allowed robots to navigate on their own over long periods and successfully cross complex terrains.
Driving through unpredictable city traffic requires constant adjustment. If you only look at the car directly in front, you might brake too late or miss a turn. In this paper, the robot uses its camera and internal sensors to anticipate obstacles and plan movements ahead, similar to how a driver uses mirrors and road signs to navigate safely.
This advances autonomous robot navigation in complex environments, reducing the need for human intervention in tasks like exploration or traversal of rough terrain.
The abstract does not explicitly state limitations, but the work is demonstrated only for quadruped robots in parkour navigation, so its generalizability to other robot types or simpler terrains is not established.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
MulDP: Multimodal Diffusion Policy for Autonomous Quadruped Parkour Navigation across Complex Terrains — Kangmai Hu, Yueqi Zhang, Peng Zhai et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.