CALM

Comfort-Aware Longitudinal Motion

Slow down for the bump, not for the sake of it.

Every driver knows the moment. A speed bump comes up, you have half a second to decide, and the car answers with a jolt that should have been a ripple. CALM is an open research project about that decision: how a car can read the road ahead and ease off by exactly as much as the bump deserves, so the people inside barely notice it.

CALM began as a bachelor thesis at TU Berlin and has grown into an open source stack for comfort-aware speed recommendation. It contributes:

An open source software stack for the slowdown-depth decision: a half-car simulator with nonlinear suspension, a Gymnasium environment, an acados MPC, a PPO policy and ISO 2631-1 comfort scoring.
The CALM dataset: 159 real bump crossings from CARD (114 Germany, 45 Italy; 105 usable), an extended real-time dataset for comfort.

Models and data

CALM-PPOThe trained speed recommender, with its five training runs. CALM-ScenariosRoad profiles of 159 real crossings, ready to simulate. CALM-Germany114 crossings with camera, depth and sensor recordings. Access on request. CALM-Italy45 crossings with camera, depth and sensor recordings. Access on request.

Code

The simulator and the planners are on GitHub, together with the MATLAB ride models in ride-models-matlab.

Acknowledgements: the real-road recordings build on the CARD dataset (Elazab et al., CVPR 2026).