Abstract— The Autotaxi system is a safety critical sensor system that is being specially developed to perform the sensing required for an autonomous vehicle to drive safely along a dedicated paved guideway network and to avoid collision. Therefore, the host vehicle is equipped with a set of sensors used to detect and track any object of interest in the field of view and to describe the guideway in which the vehicle is driving. In this work a decentralised architecture referred to as sequential pair-wise track-to-track fusion is proposed to solve the multiple-sensor multiple-target tracking data fusion problem under the context of the Autotaxi system. The approach consists of four basic stages: data alignment, redundancy elimination logic, Kalman filtering with resetting and track-to-track association and fusion. A coefficients based fusion approach is proposed to give solution to the multiple sensor guideway data fusion problem. Results from the latest test trials carried out at the Cardiff test track are presented. Index Terms— Sensor fusion, Multiple sensor multiple target tracking, guideway data fusion, Autonomous vehicles. I. INTRODUCTION The University of Bristol is participating in a collaborative project to develop a safety critical sensor system, referred to as ‘Autotaxi’, to perform the sensing required by small autonomous electric vehicles, referred to as Urban Light Transport (ULTra) vehicles. These vehicles are part of a new urban personal rapid transport (PRT) system specially designed for short-range low-speed passenger transportation based on a dedicated paved guideway network [1]. The guideway network on which the vehicles operate has inherent safety features: it is one way, it is physically segregated from other traffic and pedestrians and it is bounded on either side by kerbs capable of containing the vehicle if necessary. In order to drive safely along the guideway network an ULTra vehicle must detect the presence of obstacles, including other vehicles on the path ahead and on merging paths at junctions. Thus, two are the main tasks that Autotaxi is intended to complete: • Provide the vehicle’s Collision Avoidance System with high integrity data describing the location and The authors are with the Department of Aerospace Engineering, University of Bristol, Queens Building, University Walk, Bristol, BS8 1TR, UK (phone: +44 117 928 7704; fax: +44117 927 2771; e-mail: J.Escamilla@bristol.ac.uk; nick.lieven@bristol.ac.uk). trajectory of other vehicles and stationary obstacles on the guideway, including vehicles on merging paths at junctions, with which the host vehicle might potentially collide. • Verify given data (digital road map) describing the road ahead of the host vehicle and the vehicle’s motion on it. The sensor fusion approach is one of the most critical problems in the overall PRT project because of the safety issues involved both for the occupants of the ULTra vehicles and for the other roadway users (other vehicles, pedestrians). A. Outline System Solution The structure of the proposed outline system solution for the Autotaxi system is divided in three parts: general multi- sensor data fusion (MSDF) architecture, prototype sensor suit solution, and generic MSDF model. The MSDF architecture used in the Autotaxi system has been recently developed by one of the project partners (TRW Conekt) for the CARSENSE project [2], which is a MSDF system to detect obstacles in front of a host car. This architecture, shown in Fig. 1, is based on intelligent sensors with the aim to allow almost any sensor to be integrated into the system in a ‘plug and play’ manner [3]. This makes the architecture flexible, modular and platform independent. The term “intelligent sensors” is used to refer the kind of sensors able to provide their own data processing and object tracking streams. This reduces the amount of data to be transmitted to the central fusion processor, reducing the bandwidth requirements of communications and reducing cost by allowing relatively low speed networks to be used. In the context of the Autotaxi system an investigation has identified vision, lidar, radar, and ultrasonic as the most adequate sensor technologies for the task of obstacle detection and guideway/path description. Therefore, a sensor suit solution for the Autotaxi system considering these technologies has been recommended as is shown in Fig. 2. This sensor suit has been determined by comparing different sensing technologies for obstacle and guideway detection. It is expected that the proposed sensor suit will serve as a prototype to enable a more detailed assessment of the sensors capabilities and limitations for this particular Sensor Fusion Approaches to Guideway and Obstacle Detection in the Autotaxi System P. J. Escamilla-Ambrosio and N. Lieven 0-7803-9286-8/05/$20.00 © 2005 IEEE