Abstract

This thesis presents new tools and methods for the design and validation of advanced driver assistance systems (ADASs). ADASs aim to improve driving comfort and traffic safety by assisting the driver in recognizing and reacting to potentially dangerous traffic situations. A major challenge in designing these systems is to guarantee high performance and dependability under all possible combinations of traffic scenarios, operating conditions, and failure modes. These stringent requirements necessitate fault-tolerant control techniques and a thorough validation of the system. A microscopic traffic simulation within the simulation environment PreScan supports the initial system design. In addition, a unique tool for the design and validation of ADASs is presented and evaluated: vehicle hardware-in-the-loop (VeHIL) simulation. The VeHIL laboratory allows an ADAS-equipped vehicle to be tested in an artificial environment, where surrounding traffic is emulated by robot vehicles. VeHIL enables repeatable, safe, and accurate testing, complementary to human-in-the-loop test drives. The use of these three tools (PreScan, VeHIL, and test drives) is combined in a methodology for probabilistic validation of ADASs, based on randomized algorithms. This methodology is more efficient than conventional simulation techniques and the current practice of trial-and-error test drives. It results in a test schedule definition with a minimum number of simulations and test runs, such that the performance and dependability of an ADAS can be guaranteed, given a desired level of accuracy and confidence. The added value of the methodology is demonstrated with three case studies, involving a driver information and warning system, a fault-tolerant system for cooperative adaptive cruise control, and a pre-crash system.

Abstract

We present an approach for validation of advanced driver assistance systems, based on randomized algorithms. The new method consists of an iterative randomized simulation using adaptive importance sampling. The randomized algorithm is more efficient than conventional simulation techniques. The importance sampling pdf is estimated by a kernel density estimate, based on the results from the previous iteration. The concept is illustrated with a simple adaptive cruise control problem.

Abstract

We present a methodological approach for validation of advanced driver assistance systems, based on randomized algorithms. The new methodology is more efficient than conventional validation by simulations and field tests, especially with increasing system complexity. The methodology consists of first specifying the perturbation set and performance criteria. Then a minimum required number of samples and a relevant sampling space is selected. Next an iterative randomized simulation is executed, followed by validation with hardware tests. The concept is illustrated with a simple adaptive cruise control problem.