Autonomous vehicles control in the VisLab Intercontinental Autonomous Challenge

https://doi.org/10.1016/j.arcontrol.2012.03.012Get rights and content

Abstract

Autonomous driving is one of the most interesting fields of research, with a number of important applications, like agricultural, military and, most significantly, safety. This paper addresses the problem of designing a general purpose path planner and its associated low level control for autonomous vehicles operating in unknown environments. Different kinds of inputs, like the results of obstacle detection, ditch localization, lane detection, and global path planning information are merged together using potential fields to build a representation of the environment in real-time; kinematically feasible trajectories, based on vehicle dynamics, are generated on a cost map. This approach demonstrated both flexibility and reliability for vehicle driving in very different environments, including extreme road conditions. This controller was extensively tested during VIAC, the VisLab Intercontinental Autonomous Challenge, a 13,000 km long test for intelligent vehicle applications. The results, collected during the development stage and the experiment itself, are presented in the final part of this article.

Introduction

The VisLab Intercontinental Autonomous Challenge (VIAC)1 has been a unique chance to tackle the problem of designing a local path planner capable of dealing with multiple sources of global planning data, such as

  • digital maps;

  • waypoints provided on a radio link by a leader vehicle;

  • waypoints generated by a leader follower applications.

While also exploiting information on the immediate vehicle surroundings, including

  • obstacles detected by LIDARs;

  • obstacles detected by the stereo vision systems;

  • lane markings;

  • ground roughness and ditches;

  • position and trajectory of the preceding vehicle, if any.

Local path planning still represents one of the major areas of research in robots motion planning, and is usually carried out by using one of the following approaches:

  • Sliding Mode Path Following (SMPF), where the steering control law is based on the error between the current position/orientation and a given feasible trajectory;

  • Traversability-Anchored Dynamic Path Following (TADPF), where starting from the current position all feasible trajectories are evaluated and the safest one is determined.

Both strategies can nevertheless be combined in real-time in order to boost their strengths, as suggested by Maček, Philippsen, and Siegwart (2009).

During the 2005 DARPA Grand Challenge the TerraMax2 vehicle was able to negotiate 220 miles of offroad driving with no human intervention using a variant of RTT3 (Braid, Broggi, & Schmiedel, 2006), a common approach to select the most promising trajectories: up to 2000 different trajectories were evaluated for each planning cycle to determine the best path. TerraMax (Chen et al., 2008) in the DARPA Urban Challenge implemented different state-of-the-art trajectory generators that were triggered by a high level behavior supervisor, depending on both the tasks that had to be executed and environmental conditions.

Stanley, the Stanford Racing Team’s entry in the DARPA Grand Challenge (Hoffmann, Tomlin, Montemerlo, & Thrun, 2007), presented a new approach in tracking the generated trajectory based on a lateral error estimation and considering the front wheels orientation. The trajectory was required to be smooth and match feasible curvatures (Thrun et al., 2006).

During the DARPA Urban Challenge Junior (Dolgov, Thrun, Montemerlo, & Diebel, 2010) implemented a mid-term pathplanner which generated trajectories on a low resolution grid using a hybrid A* algorithm, considering both obstacles and vehicle constraints. As a second step, this optimal trajectory was smoothed using a Conjugate Gradient technique on a higher resolution Voronoi field. An important issue discussed in that paper is the trade-off between keeping obstacles at a safe distance and the cost of extending the route. Finally the smooth trajectory that was generated was followed using a SMPF technique.

Ben Franklin Racing Team’s (Bohren et al., 2008) used a similar formula based on lateral error to control the steering angle.

Caltech’s Alice (Linderoth, Soltesz, & Murray, 2008) implemented the generation of optimal trajectories using a non-linear optimization of the cost function. The initial rough trajectories were optimized at a later stage and the generated path was followed by a lateral control error tracking. If the error exceeded a given threshold, the vehicle invoked a replanning phase.

CMU’s Boss (Urmson et al., 2008, Urmson et al., 2009) used a model-predictive trajectory generator, the same proposed by Howard, Green, Kelly, and Ferguson (2008), to produce dynamically feasible actions between the initial and the desired vehicle states using numerical linearization and inversion of the forward vehicle dynamic model.

Talos (Kuwata et al., 2009) implemented a variant of RTT: the feasibility of the proposed trajectories was evaluated on a drivability grid, where each cell stored a drivable/non-drivable flag and the cost to drive over it.

In more recent times VisLab designed and built BRAiVE (see Fig. 1), a mobile laboratory for the development and validation of control and machine vision technologies. The sensors suite comprises 10 cameras, 5 laser scanners, 1 radar, 1 GPS+IMU and 1 E-Stop (emergency stop) system.

Redundancy is intentional and enables the use of different subsets of sensors when developing Advanced Driver Assistance Systems (ADAS) applications (e.g. pedestrian detection, obstacle detection, vehicle following, collision warning, traffic sign recognition, parking slot detection, backup maneuver, collision detection, etc.).

A shown in Fig. 2, a dSpace MicroAutoBox unit acts as a gateway between actuators and BRAiVE autonomous driving system. This computer translates the CAN messages coming from the Control PCs into another stream, suitable for actuators, and stops the car when it receives the emergency signal or when it detects a fault condition.

All these robots show a clear division between the path planner and the low level steering control, having to perform global tasks which require a medium-term planning. In fact they demonstrate that the use of a global planner allows to execute complex maneuvers such as parking and U-turns.

To complete the VIAC expedition, the vehicles had to be able to run relying on a local planner only, that in the future could be extended to a global path planner. Methods to evaluate trajectories based on potential field are common in the literature, allowing a modular development of sensors. For this reason the trajectory is generated from the evaluation of a subset of possible acceptable vehicle trajectories on a cost map. An optimal trajectory is provided at a high rate in order to cope with the unstable nature of vehicle dynamics and with rapid environmental changes.

The remainder of this paper is organized as follows. In Section 2 the electric vehicles involved in expedition are presented and in Section 3 the equations underlying the control system are explained. Since proper knowledge of vehicle dynamics is essential, a calibration phase is also described in this paper. The control system is described in Section 4 and finally, the results of the tests and the whole experiment, based on a large amount of kilometers of autonomous driving, are presented in Section 5.

Section snippets

Vehicle setup

Thanks to the cooperation with Piaggio, the electric vehicles selected for VIAC are Piaggio Porter Electric Power vans (Fig. 3).

To control vehicle speed and steering, different devices have been installed:

  • A servo motor (Fig. 4), provided by TopCon, is directly connected to the steering wheel column and controlled in position, speed, and torque through CAN-BUS messages. The set-point is internally followed using a PID controller.

  • Speed is adjusted by changing the duty cycle of the PWM signal that

Kinematic model

Due to low vehicle speeds, a kinematic model has been chosen over the dynamic one. This section shows how it was extended to handle also a higher speed steady-state curve condition.

Since vehicles are rigid bodies, they can be modelled as oriented points moving on a plane. In a kinematic model this point is located on the rear axle. This configuration is defined by position (x, y) and orientation θ in an absolute reference frame.

The motion model (shown in Fig. 8) is described by the kinematic

Vehicle Control Design

In order to provide reliable steering, throttle, and brake control, the pathplanner considers the different inputs obtained by cameras and laser processing, GPS/INS, and the high level control interfaces, as shown in Fig. 11. The kinematic model is determined according to these information, and provides the speed v and the curvature κ that are used as the set points for the low-level control sub-systems.

Conclusions

This paper presented the control system developed for VIAC and used in other autonomous driving experiments.

The most important feature of the controller consists in limiting the number of evaluated trajectories, generating only candidates that the underlying control system can handle. The use of cost maps allows fusion of information provided by several sensors in a very simple and effective way providing a fast assessment of the trajectories reliability. Therefore the planner is capable of

Acknowledgments

The work described in this paper has been developed in the framework of the Open intelligent systems for Future Autonomous Vehicles (OFAV) Project funded by the European Research Council (ERC) within an Advanced Investigators Grant.

Alberto Broggi received the Dr. Ing. (Master) degree in Electronic Engineering and the Ph.D. degree in Information Technology both from the Università di Parma, Italy, in 1990 and 1994, respectively. He is now Full Professor at the Università di Parma. He is the Director of the Artificial Vision and Intelligent Systems Lab, VisLab, and author of more than 150 publications on international scientific journals, book chapters, refereed conference proceedings. He served as Editor-in-Chief of the

References (19)

  • J. Bohren et al.

    Little ben: The ben franklin racing team’s entry in the 2007 darpa urban challenge

    Journal of Field Robotics

    (2008)
  • D. Braid et al.

    The terramax autonomous vehicle: Field reports

    Journal of Robotics Systems

    (2006)
  • Y.-L. Chen et al.

    TerramaxTM: Team Oshkosh urban robot

    Journal of Field Robotics

    (2008)
  • D. Dolgov et al.

    Path planning for autonomous vehicles in unknown semi-structured environments

    International Journal of Robotics Research

    (2010)
  • J.C. Gerdes et al.

    A unified approach to driver assistance systems based on artificial potential fields

    Journal of Dynamic Systems, Measurement, and Control

    (2001)
  • T.D. Gillespie

    Fundamentals of Vehicle Dynamics

    (1992)
  • Hoffmann, G., Tomlin, C., Montemerlo, M., & Thrun, S. (2007). Autonomous automobile trajectory tracking for off-road...
  • T. Howard et al.

    Optimal rough terrain trajectory generation for wheeled mobile robots

    International Journal of Robotics Research

    (2007)
  • T.M. Howard et al.

    State space sampling of feasible motions for high-performance mobile robot navigation in complex environments

    Journal of Field Robotics

    (2008)
There are more references available in the full text version of this article.

Cited by (112)

  • Recent advances in motion and behavior planning techniques for software architecture of autonomous vehicles: A state-of-the-art survey

    2021, Engineering Applications of Artificial Intelligence
    Citation Excerpt :

    However, because of turning of the steering wheel, curvature variation is introduced in the motion of vehicle and vehicle motion can be approximated as a clothoid path, as shown in Fig. 12. Broggi et al. (2012) used clothoids to represent the path and trajectory is generated from the evaluation of a subset of possible acceptable vehicle trajectories on a cost map. A new trajectory based on clothoids is also used, which joins the previously planned trajectory to avoid obstacles (Bertolazzi et al., 2018).

  • A survey of autonomous vehicles for traffic analysis

    2024, Indonesian Journal of Electrical Engineering and Computer Science
  • Review of Research on Decision-making and Planning for Automated Vehicles

    2024, Zhongguo Gonglu Xuebao/China Journal of Highway and Transport
View all citing articles on Scopus

Alberto Broggi received the Dr. Ing. (Master) degree in Electronic Engineering and the Ph.D. degree in Information Technology both from the Università di Parma, Italy, in 1990 and 1994, respectively. He is now Full Professor at the Università di Parma. He is the Director of the Artificial Vision and Intelligent Systems Lab, VisLab, and author of more than 150 publications on international scientific journals, book chapters, refereed conference proceedings. He served as Editor-in-Chief of the IEEE Transactions on Intelligent Transportation Systems for the term 2004–2008; he served the IEEE Intelligent Transportation Systems Society as President for the term 2010–2011, and now he’s holding the Past-President position.

Paolo Medici, born in Reggio Emilia, Italy, holds a Ph.D. in Information Technology and a MSc in Electronic Engineering, both from the Università di Parma, Italy. After graduating in 2004 he began working as researcher in the Artificial Vision and Intelligent System Laboratory (Vislab) with the Dipartimento di Ingegneria dell’Informazione, Università di Parma. His researches embrace many fields of computer vision, sensors calibration techniques, ensemble learning and vehicle control for the development of advanced driver assistance systems and future autonomous vehicles.

Paolo Zani received the MSc degree in Computer Engineering from the Università di Parma, Parma, Italy, in 2005. In 2009, he received the Ph.D. degree in Information Technology from the Università di Parma. His research activity is focused on efficient computer vision algorithms for intelligent and autonomous vehicles. He participated to the 2005 DARPA and 2007 DARPA Challenges, and to the VisLab Intercontinental Autonomous Challenge as X-by-wire and lane detection systems developer.

Alessandro Coati obtained his BSc and MSc degrees in Information Engineering from Università degli Studi di Parma, Italy, in 2005 and 2009, respectively. From 2009, he is working as a researcher at the VisLab, the artificial vision laboratory of Parma University, involved in several projects about AGV and obstacle detection for industrial environment. Since 2011, he is a Ph.D. student in “Information Technologies” from the Department of Information Engineering of the Università di Parma.

Matteo Panciroli was born in Castelnovo ne’ Monti, Reggio Emilia, Italy, on December 29, 1984. He received – “summa cum laude” – the MSc degree in Information Engineering, with a thesis on “Progettazione e sviluppo di un sistema di percezione tramite fusione sensoriale per ACC” (“Design and development of a perception system based on sensor fusion for ACC”) from the Università di Parma, Parma, Italy, in July, 2008. Since January 2009, he is a Ph.D. student in “Information Technologies” from the Department of Information Engineering (DII) of the Università di Parma. In 2010 he has been in charge of sensor calibration task during VIAC project. His research activity field is on autonomous vehicle and advanced driver assistance systems and it is focused on sensor calibration and 3d vision applications.

An earlier version of this paper appeared in Broggi, A., et.al. ‘Development of the control system for the VisLab Intercontinental Autonomous Challenge’, Intelligent Transportation Systems (ITSC), Madeira, Portugal, October 2010.

View full text