Stochastic dynamic programming control policies for fuel efficient vehicle following

Abstract

Stochastic control policies are developed for fuel-efficient following of a lead vehicle moving at a constant speed. Assuming that a lead vehicle is travelling at a specified speed and that the road grade is modeled stochastically, we solve numerically a Stochastic Dynamic Programming (SDP) problem to determine follower vehicle speed control policies subject to constraints on vehicle following distance. The cost function to which SDP is applied reflects fuel consumption, travel time and includes a penalty for distance constraint violation. A virtual testing environment based on CarSim is used for simulations and fuel economy assessment. We demonstrate the potential for fuel consumption improvements (over 5%) and examine properties of the speed control policy of the follower vehicle, including the emergent `pulse and glide' behavior. The policies do not require preview information of the route and of the road grade thereby providing a potentially simpler alternative to preview-based systems. The results complement our previous results on fuel efficient in traffic driving.