[Paper] Real-time Safety-critical NMPC with Control Barrier Function (CBF)

To achieve safety and comfort, autonomous vehicles often use prediction horizon to optimize control and motion. With limited real-time control computation, this horizon is also limited, sometimes too short to properly avoid collision or make comfortable braking/lane change.

4-5 years ago we developed an intuition to deal with this challenge, learning from control barrier function (CBF) from legged robotics community. Inspired by Quan Nguyen (MIT, now Prof. in USC), we obtained some positive results, and presented in CDC19 (https://lnkd.in/ecd4QbZ4). CBF constraint indeed helps to extend the prediction horizon implicitly (theoretically infinitely), making the car drives more safe and comfortable.

Recently, our industrial PhD research engineer Jean Pierre Allamaa extended and showed CBF can be incorporated into a collocation real-time MPC computationally & efficiently as well. It generates safe trajectories of states and inputs as splines, satisfying constraints over prediction horizon through manipulation of the splines’ coefficients.

The works are in collaboration with Prof. Toshiyuki Ohtsuka (Univ. Kyoto) and Prof. Panagiotis (Panos) Patrinos (KULeuven). Below is a small video demo to validate and benchmark using Siemens Simcenter Amesim and Prescan. Hope you enjoy :) !

If you are interested for details, please see the paper here: https://lnkd.in/evybB3S7 . It just got to be among few papers selected for oral single-track presentation in IFAC Conference on NMPC this August.