Nonlinear Lateral Control of Vision Driven Autonomous Vehicles*

Transcription

1 Paper Machine Intelligence & Robotic Control, Vol. 5, No. 3, (2003) Nonlinear ateral Control of Vision Drien Autonomous Vehicles* Miguel Ángel Sotelo Abstract: This paper presents the results of a lateral control strategy that has been applied to the problem of steering an autonomous ehicle using ision. The lateral control law has been designed for any kind of car-like ehicle presenting the Ackerman kinematic model, accounting for the ehicle elocity as a crucial parameter for adapting the steering control response. This makes the control strategy suitable for either low or high speed ehicles. The stability of the control law has been analytically proed and experimentally tested by autonomously steering Babieca, a Citroen Berlingo prototype ehicle. Keywords: Vehicle Control, Chained Systems Theory, Vision Drien Autonomous Vehicles 1. Introduction ATERA automatic steering of autonomous car-like ehicles has become an apparent field of application for robotics researchers. Basically, this problem can be stated as that of determining an appropriate control law for commanding the ehicle steering angle. Many steering control designs are already documented in the literature [1], [3], [6]. A comparatie study on arious lateral control strategies for autonomous ehicles can be found in [11], where a linearized model of the lateral ehicle dynamics is used for controller design based on the fact that it is possible to decouple the longitudinal and lateral dynamics. On the contrary, a simplified nonlinear lateral kinematic model is proposed in this work to ease the design and implementation of a stable lateral control law for autonomous steering of car-like ehicles. The lateral control strategy was implemented on Babieca, an electric Citroen Berlingo experimental prototype, using ision as the main sensor to measure the position of the ehicle on the road. Real tests were carried out on a priate circuit, emulating an urban quarter, composed of streets, intersections (crossroads), and roundabouts, located at the Industrial Automation Institute (IAI) in Arganda del Rey, Madrid. Additionally, a lie demonstration exhibiting the system capacities on autonomous steering was carried out during the IEEE Conference on Intelligent Vehicles 2002, in a priate circuit located at Satory (Versailles), France. 2. ateral Control Considering the case of an autonomous ehicle driing along some reference trajectory, the main goal of the lateral control module is to ensure proper tracking of the reference trajectory by correctly keeping the ehicle in the center of the lane with the appropriate orientation (parallel to the desired trajectory). This constraint can be generalized as the minimization of the lateral and orientation errors of the ehicle (d e, ) with respect to the reference trajectory, at a Receied 20 August, 2003; accepted 13 October, Department of Electronics. Technical School, Uniersity of Alcalá, Campus Uniersitario s/n, Alcalá de Henares, Madrid, Spain. michael@depeca.uah.es 2. 1 Kinematic model The kinematic model of the ehicle is the starting point to model the dynamics of the lateral and orientation errors. The ehicle model is approximated by the popular Ackerman (or bicycle) model [4] as depicted in Fig. 2, assumc 2003 Cyber Scientific Fig. 2 Fig. 1 Control point θ (x d,y d ) θ d d e Reference trajectory ateral and orientation errors at the look-ahead distance Ackerman steering φ R Approximate kinematic model of the ehicle (Ackerman steering) control point, as illustrated in Fig. 1. To sole this controllability problem and design a stable lateral controller, a model describing the dynamic behaior of d e and is needed. φ Paper No X/03/

2 88 M. A. Sotelo ing that the two front wheels turn slightly differentially and thus, the instantaneous rotation center can be purely computed by kinematic means. et κ(t) denote the instantaneous curature of the trajectory described by the ehicle: κ(t) 1 tan φ(t) dθ(t) (1) R(t) ds where R is the radius of curature, is the wheelbase, φ is the steering angle, and θ stands for the ehicle orientation in a global frame of coordinates. The dynamics of θ is computed in Eq. (2) as a function of ehicle elocity : θ dθ dθ ds ds φ(t) κ(t)(t) tan (t). (2) et φ and be the ariables of the ehicle control input space. On the other hand, the ehicle configuration space is composed of the global position and orientation ariables, described by (x, y, θ), under the flat terrain assumption. Mapping from the control input space to the configuration space can be soled by using the popular Fresnel equations, which are also the so-called dead reckoning equations typically used in inertial naigation. Equation (3) shows the dynamics of (x, y, θ): ẋ dx ẏ dy θ dθ (t) cos θ(t) (t) sin θ(t) (3) (t)tan φ(t) where (t) represents the elocity of the midpoint of the rear axle of the ehicle, denoted as the control point. Global information about the position and orientation of the ehicle (x, y, θ) is then transformed so as to deelop a model that describes the open-loop lateral and orientation error dynamics. As obsered in Fig. 1, the lateral error d e is defined as the distance between the ehicle control point and the closest point along the ehicle desired trajectory, described by coordinates (x d,y d ). This implies that d e is perpendicular to the tangent to the reference trajectory at (x d,y d ). The scope of the tangent at (x d,y d ) is denoted by θ d and represents the desired orientation at that point. Based on this, d e and suffice to precisely characterize the location error between the ehicle and some gien reference trajectory, as described in Eqs. (4) and (5): d e (x x d ) sin θ d +(y y d ) cos θ d (4) θ θ d. (5) Computing the deriatie of d e with respect to time yields Eq. (6), while the time deriatie of is shown in Eq. (7): d e ẋ sin θ d +ẏ cos θ d V cos θ sin θ d + V sin θ cos θ d V sin(θ θ d ) V sin (6) d(θ θ d) θ θ d θ. (7) Thus, the complete nonlinear model for d e and is formulated in Eq. (8): d e V sin V tan φ. (8) 2. 2 Nonlinear control law The control objectie is to ensure that the ehicle will correctly track the reference trajectory. For this purpose, both the lateral error d e and the orientation error must be minimized. On the other hand, for simplicity, ehicle elocity will be assumed to be constant. The design of the control law is based on general results of the socalled chained systems theory. An excellent example on this topic can be found in [5]. Neertheless, these results are extended and generalized in this paper so as to proide a stable nonlinear control law for steering of Ackerman-like ehicles, based on local errors. From the control point of iew, the use of the popular tangent linearization approach is aoided as it is only alid locally around the configuration chosen to perform the linearization, and thus, the initial conditions may be far away from the reference trajectory. On the contrary, some state and control ariable changes are posed in order to conert the nonlinear system described in Eq. (8) into a linear one, without any approximation (exact linearization approach). Neertheless, due to the impossibility of exactly linearizing the systems describing mobile robots dynamics, these nonlinear systems are conerted into almost linear ones, termed as chained forms. The use of the chained form permits to design a control law using linear systems theory to a higher extent. In particular, the nonlinear model for d e and (Eq. (8)) can be transformed into chained form using the state diffeomorphism and the change of control ariables as in Eq. (9): y1 de Y Θ(X) y 2 tan cos (9) w1 W Υ(U) tan φ. w 2 cos 2 These transformations are inertible wheneer the ehicle speed is different from zero and the orientation error is different from π/2. This implies that the singularities of the transformations can be aoided by assuring that the ehicle moes ( >0) and that its orientation error is maintained under 90 degrees (the ehicle orientation must not be perpendicular to the reference trajectory). These conditions are reasonably simple to meet in practice. From Eq. (9), the ehicle model can be rewritten as in Eq. (10), considering y 1 and y 2 as the new state ariables: ẏ 1 d e sin w 1 y 2 ẏ 2 d(tan ) 1 cos 2 θe tan φ cos 2 w 2. (10) In order to get a elocity independent control law, the time deriatie is replaced by a deriation with respect to ς, the abscissa along the tangent to the reference trajectory

3 Nonlinear ateral Control of Vision Drien Autonomous Vehicles 89 Control point ς y 1 is zero, then y 2 is also zero (y 2 y 1). ikewise, if y 2 is zero then is zero from Eq. (9) (y 2 tan ). Thus, both ariables y 1 and y 2 tend to zero as ariable ς grows. The preious statement is analytically expressed in Eq. (16): Fig. 3 ς cos d e Reference trajectory Graphical description of ς as graphically depicted in Fig. 3. Analytically, ς is computed as the integral of elocity ς, measured along axis ς: ς ς cos ς dς cos w 1. (11) The time deriaties of the state ariables y 1 and y 2 are expressed as functions of ς in Eq. (12): ẏ 1 dy 1 dy 1 dς dς ẏ 2 dy 2 dy 2 dς y 1 ς dς y 2 ς (12) where y 1 and y 2 stand for the deriaties of y 1 and y 2 with respect to ς. Soling for y 1 and y 2 yields Eq. (13): y 2 ẏ2 ς y 1 ẏ1 ς sin tan φ cos 2 cos cos tan y 2 tan φ w 2 w cos 3 3. w 1 (13) As obsered in the preious equation, the transformed system is linear and thus, state ariables y 1 and y 2 can be regulated to zero (so as to yield d e d e,ref 0and,ref 0) by using the control low proposed in Eq. (14): lim d e lim 0. (16) ς ς Accordingly, ariable ς must always grow so as to ensure that both d e and tend to zero. This condition is met wheneer >0and π/2 < <π/2. In other words, the ehicle must continuously moe forward and the absolute alue of its orientation error should be below π/2 in order to guarantee proper trajectory tracking. Thus, the nonlinear control law is finally deried from Eqs. (13) and (14) such that φ arctan [ cos 3 (K d tan + K p d e ) ]. (17) The control law is then modified by a sigmoidal function as shown in Eq. (18) to account for the physical limitations in the ehicle wheels turning angle and preent from actuator saturation: ] φ arctan [ Kcos 3 K(Kd tan θe+kpde) 1 exp K(Kd 1 + exp tan θe+kpde) (18) On the other hand, the use of sigmoidal functions preseres the system stability [10]. The control law is saturated to φ max by properly tuning parameter K. Thus, the maximum alue of Eq. (18) is φ max ± arctan( K). Therefore, K is chosen to ensure that φ max ±π/6 rad (physical limitation of the ehicle) gien the wheelbase 2.69 m, yielding a practical alue K : K tan π 6. (19) From obseration of Eq. (15), the dynamic response of the ariable y 1 can be considered to be a second order linear one. In practice, it is not indeed linear due to the sigmoidal function used to saturate the control law, although it can be reasonably approximated as such. Thus, an analogy between constants K d, K p and the parameters of a second order linear system, ξ (damping coefficient) and ω n (natural frequency), can be established, yielding Eq. (20): w 3 K d y 2 K p y 1, (K d,k p ) R +2. (14) Using Eqs. (13) and (14) and soling for ariable y 1 yields Eq. (15): ω n K p ξ K d 2 K p. (20) y 1 + K d y 1 + K p y 1 0 (15) where the dynamic behaior of y 1 with respect to ς is proed to be linear. Under the assumption of positie alues for constants K d and K p, ariable y 1 tends to zero as long as ariable ς grows. In fact, ariable y 1 has a second order linear dynamic behaiour dominated by two poles with negatie real components. According to this, y 1 tends to zero as the independent ariable (ς in this case, not time) tends to infinite. This statement ensures that d e tends to zero as ς tends to infinite as dictated by Eq. (9) (y 1 d e ). From Eq. (13), if ikewise, system oershoot M p and settling distance d s (gien that the system error dynamics is described as a function of space ariable ς, not time) can be obtained from Eq. (21): ξπ M p exp 1 ξ 2 d s 2% 4 ξω n. (21) The design of constants K d and K p is undertaken considering that the system oershoot must not exceed 10%

4 90 M. A. Sotelo of the step input and the settling distance should be below some gien threshold. Thus, for a typical settling time t s 20s and gien a ehicle elocity, the proper settling distance can be computed as in Eq. (22): h Control point d s t s 20. (22) d e θ d The alue of K d is deried from Eqs. (20) and (21) yielding the elocity dependant expression in Eq. (23): (x d,y d ) K d 8 d s 0.4. (23) θ Reference trajectory ikewise, damping coefficient ξ is deried from Eqs. (20) and (21), as shown in Eq. (24): ξ 1 ( π ln 0.1 ) 2 +1 K d 2 K p 4 d s Kp. (24) Finally, K p is deduced from the preious equation, yielding Eq. (25): ( ) 2 ( ) K p. (25) d s The dependency of K p and K d on the ehicle elocity permits to ensure proper dynamic response. In particular, ehicle turning angle will be smooth at high speeds, therefore aoiding possible oscillations due to physical constraints in steering dynamics Extension of the control law for high speeds The nonlinear control law designed in the preious section proides stable trajectory tracking for Ackerman-like ehicles at moderate speeds (up to km/h). This makes the preious control law suitable for low speed ehicles. Howeer, experience demonstrates that tracking errors and ehicle oscillation increase as elocity rises. Then it becomes necessary to deelop an extension of the nonlinear control law for high speeds. The first step is to modify the ehicle control point as depicted in Fig. 4, in order to anticipate the trajectory curature at a gien distance h denoted by ook-ahead distance. The new lateral and orientation errors are then computed as illustrated in the same Fig. 4, yielding the results in Eq. (26): d e (x + h cos θ x d ) sin θ d +(y + h sin θ d y d ) cos θ d (26) θ θ d. The choice of h is carried out based on the current ehicle elocity, as described in [3], yielding the parameters shown in Eq. (27): min if < min h () t 1 if min max (27) max if > max where t s is the look-ahead time, min 25km/h, max 75km/h, min m, and max m. Considering the same scheme followed in the preious Fig. 4 ateral and orientation errors at the look-ahead distance section, the new nonlinear model for d e and is shown in Eq. (28): d e sin + h cos tan φ (28) tan φ. This model can be transformed into chained form using the state diffeomorphism and change of control ariables, as in Eq. (29): y1 de Y Θ(X) y 2 tan W w1 Υ(U) w 2 cos + h cos 2 tan φ sin tan φ. cos 2 (29) These transformations are inertible wheneer the ehicle speed is different from zero and the orientation error is different from π/2. From Eq. (28) the ehicle dynamic model can be rewritten as in Eq. (30), considering y 1 and y 2 as the new state ariables: ẏ 1 d e sin + h ẏ 2 d(tan ) cos tan φ w 1 y 2 tan φ (30) cos 2 w 2. 1 cos 2 θe In order to get a elocity independent control law, the time deriatie is replaced by a deriation with respect to ς, a ariable related to the abscissa along the tangent to the reference trajectory. Analytically, ς is computed according to the following expression: ( ς cos + h cos 2 tan φ sin ). (31) The time deriaties of the state ariables y 1 and y 2 are expressed as functions of ς in Eq. (32): ẏ 1 dy 1 dy 1 dς dς y 1 ς ẏ 2 dy 2 dy 2 dς dς y 2 ς (32)

5 Nonlinear ateral Control of Vision Drien Autonomous Vehicles 91 where y 1 and y 2 stand for the deriaties of y 1 and y 2 with respect to ς, respectiely. Soling for y 1 and y 2 yields Eq. (33): y 1 tan y 2 (33) y 2 tan φ cos 3 cos 4 w 3. tan φ + h sin As in the preious section, the transformed system is linear and thus, state ariables y 1 and y 2 can be regulated to zero (so as to yield d e d e,ref 0and,ref 0) by using the new control low proposed in Eq. (34): w 3 K d y 2 K p y 1 (K d,k p ) R +2. (34) Using Eqs. (33) and (34) and soling for ariable y 1 yields Eq. (35): y 1 + K d y 1 + K p y 1 0 (35) where the dynamic behaiour of y 1 with respect to ς is proed to be linear. Once again, this implies that ariables y 1 ( d e ) and y 2 ( tan ) tend to zero as ariable ς grows. The preious statement is analytically expressed in Eq. (36): lim ς d e lim ς 0. (36) Accordingly, ariable ς must always grow so as to ensure that both d e and tend to zero. This condition is met wheneer >0and π/2 < <π/2. In other words, the ehicle must continuously moe forward and the absolute alue of its orientation error should be below π/2 in order to guarantee proper trajectory tracking. Basically, stability conditions remain the same as in the preious section. Thus, the new nonlinear control law for high speeds is finally deried from Eqs. (33) and (34) such that [ sin θe cos 3 ] (K d tan + K p d e ) φ arctan. sin + h cos 4 (K d tan + K p d e ) (37) The control law is then modified by a sigmoidal function to account for the physical limitations in the ehicle wheels turning angle and preent from actuator saturation. From this point onwards, tuning of K, K d, and K p follows the same scheme deried in Eqs. (19), (23), and (25), respectiely. 3. Implementation and Results The control law for autonomous steering described in this paper was tested on the so-called Babieca prototype ehicle (an electric Citroen Berlingo), as depicted in Fig. 5. The ehicle was modified to allow for automatic elocity and steering control at a maximum speed of 90 km/h. Babieca is equipped with a color camera to proide lateral and orientation position of the ego-ehicle with regard to the center of the lane, a Pentium PC, and a set of electronic deices to proide actuation oer the accelerator and steering wheel, as well as to encode the ehicle elocity and steering angle. The color camera proides standard PA ideo Fig. 5 Babieca prototype ehicle signal at 25 Hz that is processed by a Meteor frame grabber installed on a 120 MHz Pentium running the Real Time inux operating system. The complete naigation system, implemented under Real Time inux using a pre-emptie scheduler, runs a ision-based lane tracking task for computing the lateral and orientation errors. Practical experiments were conducted on a priate circuit located at the Industrial Automation Institute in Arganda del Rey (Madrid). The circuit is composed of seeral streets, intersections, and roundabout points, trying to emulate an urban quarter. Various practical trials were conducted so as to test the alidity of the control law for different initial conditions in real circumstances. During the tests, the reference ehicle elocity is kept constant by a elocity controller. Coefficients K d and K p were calculated as a function of using Eqs. (19) and (21). Figures 6 and 7 show the transient response of the lateral and orientation errors of the ehicle for reference elocities of 20 km/h and 50 km/h respectiely. In both cases, the ehicle starts the run at an initial lateral error of about 1m and an initial orientation error in the range ±5. For illustratie purposes, Fig. 8 depicts the steering control proided by the controller during the path tracking experiment carried out at 50 km/h. As can be clearly appreciated, the steady state response of the system is satisfactory for either experiments. Thus, the lateral error is bound to ±5 cm at low speeds and ±25 cm at 50km/h, while the absolute orientation error in steady state remains below 1 in all cases. Just to gie an example on how the practical results conform to the expected alues as deried from the theoretical deelopment, let s consider the transient response of the ehicle depicted in Fig. 6 for 20km/h. Assuming a theoretical maximum oershoot of M p 10% and a settling time of t s 20s, the controller coefficients are tuned to K d and K p , according to Eqs. (19) and (21). Nonetheless, from obseration of Fig. 6 the maximum oershoot obtained in practice yields up to almost 25% for both the lateral and orientation errors, while the settling time takes some 22 s. This is mainly due to the existence of nonlinear actuator dynamics and latencies, which are not considered in the model. In spite of these slight differences with regard to the theoretical expected alues, the practical results exhibited in this section demonstrate that the nonlinear lateral control law deeloped in this work still permits to safely steer the ehicle at operational elocities. In a final trial, the results achieed in the second test for

6 Orientation error ( ) Orientation error ( ) 92 M. A. Sotelo ateral error (m) ateral error (m) time(s) Time (s) 20 km/h Time time(s) Fig. 6 Transient response of the lateral and orientation error for 20km/h ateral error (m) ateral error (m) time(s) Time (s) 50 km/h Time time(s) Fig. 7 Transient response of the lateral and orientation error for 50km/h Steering response ( 1,+1) Steering response (-1, +1) Time (s) time(s) Fig. 8 Steering normalized response (between 1 and +1) for 50 km/h 20km/h are compared to human driing at the same speed along the same trajectory. For this purpose a human drier steered the ehicle, leaing the control of the accelerator to the elocity controller in order to keep a reference speed of 20 km/h. The comparison is graphically depicted in Fig. 9. One can obsere that the human drier takes less time than the automatic controller to achiee lateral and orientation errors close to zero. On the other hand, the steady state errors are similar in both cases. Surprisingly, human driing turns out in sporadic separations from the reference trajectory up to cm, without incurring in dangerous behaior, while the automatic controller keeps the ehicle under lower lateral error alues once stabilized. Far from being an isolated fact, this circumstance was repeatedly obsered in seeral practical experiments. As the conclusion, the lateral control law deeloped in this work can reasonably be considered to be alid for driing a car-like ehicle as precisely as a human can. During the last year, Babieca ran oer hundreds of kilometers in lots of successful autonomous missions carried out along the test circuit using the nonlinear control law described in this paper. A lie demonstration exhibiting the system capacities on autonomous driing using the nonlinear control law described in this paper was carried out during the IEEE Conference on Intelligent Vehicles 2002, in a priate circuit located at Satory (Versailles), France. A complete set of ideo files demonstrating the operational performance of the control system in real test circuits (both in Arganda del Rey and in Satory) can be retrieed from ftp:// 4. Conclusions To conclude, the next key points should be remarked. First of all, the nonlinear control law described in this work has proed its empirical stability for lateral dric 2003 Cyber Scientific Machine Intelligence & Robotic Control, 5(3), (2003)

7 Orientation error ( ) Nonlinear ateral Control of Vision Drien Autonomous Vehicles 93 ateral error (m) ateral error (m) * human drier - nonlinear controller error de orientación ( ) error de orientación ( ) * human drier - nonlinear controller time(s) Time (s) 20 km/h Time time(s) Fig. 9 Comparison between automatic guidance and human driing at 20km/h ing of car-like ehicles. In fact, it has been implemented on a real commercial ehicle slightly modified so as to allow for autonomous operation and tested on two different priate circuits. A key adantage of the proposed method oer other nonlinear control techniques relies on its ability to proide a priori design of the system transient response by proper fitting of constants K d and K p, while proiding stable steady state response. Vehicle commanded actuation is taken into account by considering the current elocity in the design of the controller coefficients. This permits to adapt the steering angle as a function of driing conditions at not high energy expense from the control signal point of iew. As demonstrated in practical trials, driing precision achieed by the lateral control law is as accurate as that of a human drier under normal conditions. Nonetheless, in spite of haing achieed some promising results there is still much space for improement concerning ehicle stability and oscillations. Indeed, our current work focuses on accounting for more precise ehicle models including actuator dynamics and nonlinearities. Accordingly, a new nonlinear control law should be deeloped in an attempt to increase stability and comfortability when driing at high speed. Acknowledgments This work has been funded by the Comisión Interministerial de Ciencia y Tecnología (CICYT) through research project DPI C05-04, as well as by the generous support of the Industrial Automation Institute of the CSIC (Consejo Superior de Inestigaciones Cientificas). References [1] J. Ackermann and W. Sienel, Robust control for automated steering, in Proc. of the 1990 American Control Conference, ACC90, San Diego, CA, 1990, pp [2] A. Broggi, M. Bertozzi, A. Fascioli, and G. Conte, Automatic Vehicle Guidance: the Experience of the ARGO Autonomous Vehicle. Singapore: World Scientific, [3] R. H. Byrne, C. T. Abdallah, and P. Dorato, Experimental results in robust lateral control of highway ehicles, IEEE Control Systems Magazine, ol. 18, no. 2, pp , [4] S. Cameron and P. Prober, Adanced Guided Vehicles: Aspects of the Oxford AGV Project. Singapore: World Scientific, [5]. Cordesses, P. Martinet, B. Thuilot, and M. Berducat, GPS-based control of a land ehicle, in Proc. of IAARC/IFAC/IEEE International Symposium on Automation and Robotics in Construction IS- ARC 99, Madrid, Spain, September [6] T. Hessburg and M. Tomizuka, Fuzzy logic control for lateral ehicle guidance, IEEE Control System Magazine ol. 14, pp , August [7] J. Kosecka, R. Blasi, C. J. Taylor, and J. Malik, Vision-based lateral control of ehicles, in Proc. of Intelligent Transportation Systems Conference, Boston, [8] J. uo and P. Tsiotras, Control design for systems in chained form with bounded inputs, in Proc. of American Control Conference, Philadelphia, PA. June [9] M. A. Sotelo, F. J. Rodriguez,. Magdalena,. M. Bergasa, and. Boquete, A color ision-based lane tracking system for autonomous driing on unmarked roads, Autonomous Robots, ol. 16, pp , January [10] H. Sussmann, E. Sontag, and Y. Yang, A general result on the stabilization of linear systems using bounded controls, IEEE Trans. on Automatic Control, ol. 39, no. 12, pp , January [11] C. J. Taylor, J. Kosecka, R. Blasi, and J. Malik, A comparatie study of ision-based lateral control strategies for autonomous highway driing, The International Journal of Robotic Research, ol. 18, no. 5, pp , Biography Miguel Ángel Sotelo receied the Dr. Ing. degree in Electrical Engineering in 1996 from the Technical Uniersity of Madrid, and the Ph.D. degree in Electrical Engineering in 2001 from the Uniersity of Alcalá, Alcalá de Henares, Madrid, Spain. From 1993 to 1994 he has been a Researcher at the Department of Electronics, Uniersity of Alcala, where he is currently an Associate Professor. His research interests include Real-time Computer Vision, Control Applications, and Intelligent Transportation Systems for autonomous and assisted ehicles. He has been recipient of the Best Research Award in the domain of Automotie and Vehicle Applications in Spain, in He is the author of more than 50 refereed publications in international journals, book chapters, and conference proceedings. He is an actie reiewer of seeral journals or transactions concerning Robotics, Control, and Intelligent Transportation Systems. Professor Sotelo is member of IEEE.