Decision Making and control for automotive safety Mohammad Ali

Financial supporters

Outline Motivation, background and challenges Some Suggested approaches and results Concluding remarks

Road Injuries Are among the top three causes of death for people 5-44 years old Cost governments 1-3 % of their GDP Cause US$ 518 billion in global losses Source: WHO, ”Global status report on road safety: time for action”, Geneva, 2009

Other Collision avoidance systems Aim at avoiding rear end collisions Brake when it is no longer possible to avoid colliding by braking by steering Don’t interfere unless it’s neccessary Often only mitigate Minimal nuisance while providing benefit when possible

Wait until it’s unavoidable?

Approaches and results

objective Utilize knowledge of road geometry to: Avoid loss of control Keep the vehicle on the road Without disturbing the driver

Advanced sensing Threat assessment Decision making Path planning challenges Vehicle control

Threat assessment problem Vehicle on the road and in the ”linear region”

Threat assessment problem ≤ Slip limit ≤ Half the lane width |Slip angle front | | Slip angle rear | | Deviation centerline vehicle corner 1 | | Deviation centerline vehicle corner 2 | | Deviation centerline vehicle corner 3 | | Deviation centerline vehicle corner 4 |

Threat assessment problem NowLater Admissible set Given estimates of vehicle state and surrounding environment, can we find an admissible sequence of control signals s.t. the vehicle state evolves within the prescribed constraints? Intervene Dont intervene

Reachability based approach NowLater Admissible set 1 Select terminal target set 2 Compute sequence of safe sets 3 Check whether Safe set

results Are we overestimating the driver’s capability? Safe set Admissible set Safe set Admissible set

Normal drivingRough driving Steering angle Position error Orientation error We can estimate the two gains and the look ahead time! Driver model

Results Safe set (driver model) Safe set (no driver model) Admissible set By accounting for driver limitations we can intervene earlier

Braking interventions on Braking interventions off Can we make a difference? results

results On Off

results On Off

results Braking interventions off

results Braking interventions on

results Steering interventions on

NowLater Admissible set Safe set Models and estimates are always subject to uncertainty, we can account for: Uncertainty in state estimates Uncertainty in estimates of surrounding environment (e.g. curvature, friction..) Uncertainty in model parameters (e.g. driver model parameters) How about uncertainty?

Concluding remarks

Threat assessment algorithms –papers 1-4 Driver model estimation –paper 2 Uncertainty in estimates of state, additive disturbances, model parameters –paper 3 Nonlinear dynamics –papers 1 & 4 Decision making –paper 1 Intervention design –paper 1 Everything validated through experiments –papers 1-4 contributions

Safety feature that utilizes knowledge of the road geometry to: Avoid loss of control Keep the vehicle on the road Make the vehicle easier to maneuver Driver skills are not limiting Friction estimation is difficult in low excitation Curvature estimation is difficult on bad roads Driving application Vehicle dynamics control Collision avoidance systems

Thank you for listening! Don’t run off the road like I did!