# Added vision-based foothold selection approach to my open-source MPC/WBC quadruped controller — retained record

> Retained reddit record from reddit:robotics.

Canonical: https://highsignal.app/data/records/02f15b37f8d11899

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:robotics)  published 2026-10-20 18:22 ingested 2026-10-21 02:40

[original source ↗](https://www.reddit.com/r/robotics/comments/1xazx0m/added_visionbased_foothold_selection_approach_to/)

## retained text



```
Hi everyone! I’d like to share an update to my locomotion controller for a quadruped robot I designed myself. For now, the video shows it in MuJoCo simulation.

The control loop can now use depth camera images to build a height map. The algorithm is based on D. Kim’s work [Vision Aided Dynamic Exploration of Unstructured Terrain with a Small-Scale Quadruped Robot](https://dhkim0821.github.io/papers/2020_ICRA_Vision.pdf). Using this height map, the footstep planner selects safe footholds during locomotion. This makes the robot more stable and allows it to overcome more challenging obstacles. Moreover algorithm now is able to adapt the stride height dynamically to the height of obstacles. So when walking on flat ground, the robot doesn’t need to lift its legs high, which saves energy and makes the motion look more natural.

Stack: C++, Pinocchio, MuJoCo, ROS 2, MPC + WBC

For now, the controller assumes rigid, non-slippery terrain. As a next step, I plan to improve the reference generator to further increase stability and terrain traversability.

Here’s the link to my repository: [https://github.com/voltdog/mors\_quadruped](https://github.com/voltdog/mors_quadruped). I’d be really glad if you find it interesting.

If you’re developing your own locomotion controllers, please share your experience and tell about your control approaches.
```



document fetched 2026-10-21 00:37 · canonical https://reddit.com/r/robotics/comments/1xazx0m/added_visionbased_foothold_selection_approach_to
