Driving Assist System for Ecological Driving Using Model Predictive Control

1. SICE 9th Annual Conference on Control System Driving Assist System for Ecological Driving Using Model Predictive Control Presented byー M.A.S. Kamal - Fukuoka IST Co-Authorsー Maskazu Mukai - Kyushu University Junichi Murata - Kyushu University Taketoshi Kawabe - Kyushu University Hiroshima, March 05, 2009

2. Outline • Ecological Driving • Needs of an Assist System • Modeling of the System • Controlling a Vehicle • Simulation Results • Conclusions and Future Work

3. Ecological Driving Scarce of Oil, Global warming, Environment pollutions force us to an idea of Eco-Driving. Eco Driving Aims in: • Increase in Mileage. • Reduce Emissions of CO2. • Reduce noise pollution. • Reduce accidents. • Reduce impact on environment. • Environment Friendly System

4. Realization of Eco-Driving By Proper • Vehicle maintenance, • Route Selections, • Traffic Signaling Systems, • Driving Style  Vehicle Control. A good Driving or Vehicle Control Style may save fuel consumption significantly.

5. Eco-Vehicle Control Strategy Desired behavior includes: • Minimize acceleration and braking. • Smooth and higher acceleration at starting. • Cruise at the best economy speed. • Stop by coasting or little braking. Inevitable Situations: • At urban traffic or in traffic congestion. • A red signal. • Braking preceding vehicle.

6. Conventional Eco-Strategy Remarkable feature • Eco Driving Tips based on Vehicle Engine fuel consumption characteristics. • Qualitative Assistance without rigorous reasoning: “do not accelerate very hard” Limitations: • No indication what should be the exact acceleration; no a quantitative value (e.g. 2.3 m/s2). • No analysis of the traffic trends, which greatly influenced on acceleration/braking on urban traffic. Solutions ?

7. Concept of the Proposed Eco-Strategy “Slow-and-go is always better than stop-and-go” Control the Vehicle by Anticipating future situations. A Driving Assist System can help a driver to attain or refine his Eco-Driving Skills. • Driving Assist System: • Generation of optimal action using model predictive control. • Assistance to the driver through human interface.

8. ホスト Problem Formulation Scope and Assumptions • Single Lane. • Only the immediate Preceding Car. • Flat road, no slope. • Longitudinal Motion Control. • Preceding Vehicle will move at its current acceleration/deceleration. • A Dummy vehicle stopped at red signal.

9. H HV position HV Speed PV position PV Speed P１ Problem Formulation Input： Assumption： time dependent variable, at t, remains constant for a while.

10. Model Predictive Vehicle Control • Model of Vehicle Control System • Performance Index • Optimization of Control Inputs u1 ∆u u0 uN-1 x(t) u(t) 0 1 N Sensors HV PV Model Predictive Control The problem is discretized in N step Prediction control Optimize [u0,u1,…uN-1]T to minimize the cost:

11. A continuous function for approximation of fuel consumption is derived as: Fuel Consumption Model

12. Safe Clearance Fuel Economy Reference Speed* Performance Index Dynamic weights w1, w2, w3 focus their relative contextual merits.

13. Optimization of control inputs The Hamiltonian Function is given by : Continuation method combined with Gneralized Minimum Residual Algorithm is used to derive the solution with a given initial value vector. Required condition in finding the optimum control inputs:

14. Simulations • Prediction Horizon: T= 10 sec, N=10, and h=1.0. • Simulation step 0.1 sec. • Control input constraint: -2.75u2.75 • Time headway in car following hd=1.3sec. • Parameters of a Ford Feista Car. Observation 1: Starting from Standstill with no Preceding Vehicle. • Highest initial acceleration. • Reaches full speed at about 10 sec. • Continues cruising at the best economy speed.

15. 13 m PV HV Simulations Observation 2: A typical starting situation • Both Vehicle start from standstill. • HV Started with higher acceleration. • Control input is adjusted without any braking at closing range.

16. Test Environment AIMSUNMicroscopic Traffic Simulator Functions can be Extended through API Routine to control a car in a special way

17. Simulations Application Program Interface Model Predictive Control Interface Interface Routine Info of the Host and surrounding vehicles Control Fix a car as Host Vehicle Observation 3: Comparison with Gipps based method, a default control system in AIMSUN. Fuel consumption ml and Mileages km/l are monitored.

18. 90 Vh/h Test Route 585 Vh/h 465 Vh/h 510 m 285 m 728 m 90 Vh/h Simulations • A two lane road section. • Lane changes are not controlled. • A car is forced to stop at the beginning. • Then it controlled by MPC and Gipps methods in separate run.

19. 90 Vh/h Test Route 585 Vh/h 465 Vh/h 510 m 285 m 728 m 90 Vh/h Simulations • Average fuel consumption: Gipps : 66.35 ml; MPC: 59.96 ml. Savings :6.39 ml or 10.65% • Average Millage/ economy : Gipps : 9.84 km/l; MPC: 10.96 km/l. Improvement : 11.39%

20. Conclusions • A novel concept of Eco-Driving Assist System using Model predictive control has been presented. • Vehicle control assistance is based on both Fuel consumption and traffic trends. • The algorithm has been tested in AIMSUN, a traffic simulator with pseudo-realistic environment. • In a single road section of 700m, about 11.39% improvement in Mileage.

21. Future Work • Refinement of the system to cope additional situations such as: • Roads with up-down slopes. • With known signal timing a priori. • Experimenting on real road systems. Thank you