Robotics Summer School — Autonomous Navigation
Winner of the robot competition at the 2026 ETH Robotics Summer School — long-range navigation with FAR Planner on a quadruped robot, as part of team "38° of Freedom".
Team project at the Robotics Summer School at ETH Zurich (2026), as part of team 38° of Freedom — winner of the robot competition 🏆.
Our team built an autonomous navigation stack for a quadruped robot: LiDAR-inertial state estimation (DLIO, later RESPLE for stable and accurate pose estimates), terrain analysis, and long-range path planning.
My part was the long-range planner, FAR Planner — analyzing its failure modes and tuning it until it navigated our challenge scenarios reliably:
- Diagnosed known issues: the visibility graph is short-lived, the planner can get stuck in corners, and increasing pointcloud retention time worsens performance.
- Tuned the planner in two stages — first in simulation on representative maze scenarios, then online on the robot — together with smoothing the terrain analysis (larger voxel size, longer update time for temporal consistency).
Navigation result: the tuned FAR Planner guiding the quadruped through a maze scenario — MuJoCo simulation (left) and the visibility graph and pointcloud in RViz (right).
One of the representative simulation scenarios used for planner tuning.
Team 38° of Freedom with our winner certificates from the 2026 ETH Robotics Summer School robot competition.