Pavel Pochobradský presents Geometric Model Predictive Path Integral for Agile UAV Control With Online Collision Avoidance

We invite you to attend the upcoming seminar presenting our recent paper
"Geometric Model Predictive Path Integral for Agile UAV Control With Online
Collision Avoidance," published in IEEE RA-L. This work introduces Geometric
Model Predictive Path Integral, a novel sampling-based controller that enables
Uncrewed Aerial Vehicles to track agile trajectories while avoiding obstacles
in
real-time. Addressing the classical trade-off between tracking precision and
collision avoidance, the method leverages geometric SE(3) control to generate
precise rollout trajectories alongside random exploratory ones, and
incorporates
depth image data directly into the control loop — a first for such systems.
The results demonstrate remarkable capabilities: achieving tracking accuracy
comparable to obstacle-blind geometric controllers while surpassing
state-of-the-art planners and learning-based methods in collision avoidance,
with real-world experiments validating flight speeds up to 17 m/s and reliable
obstacle avoidance at 10 m/s. Join us to explore how this unified approach
eliminates the latency issues of traditional modular navigation systems while
pushing the boundaries of agile autonomous flight.

The seminar will be held in a short format of 20 minutes + 10 minutes Q&A.

Reference:
P. Pochobradský, O. Procházka, R. Pěnička, V. Vonásek and M. Saska,
"Geometric Model Predictive Path Integral for Agile UAV Control With Online
Collision Avoidance," in IEEE Robotics and Automation Letters, vol. 11, no. 5,
pp. 5334-5341, May 2026, doi: 10.1109/LRA.2026.3668528.

Paper: https://doi.org/10.1109/LRA.2026.3668528
Video: https://youtu.be/HEo4MQNX6xc
Code: https://github.com/ctu-mrs/gmppi