This action might not be possible to undo. Are you sure you want to continue?

# IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 18, NO.

2, MARCH 2010

469

**Decentralized Spacing Control of a String of Multiple Vehicles Over Lossy Datalinks
**

´ Rodney Teo, Member, IEEE, Duˇan M. Stipanovic, Member, IEEE, and Claire J. Tomlin, Senior Member, IEEE s

less datalinks. Wireless datalinks are however not perfect and suffer from delays and dropouts. Sheikholeslam and Desoer [16] studied the performance of the platoon when there is continually no communication of lead vehicle information. It was shown that string stability is lost without information of the lead vehicle. Hedrick et al. [17] studied the effects of communication delays from the lead vehicle on string stability. They concluded that any delay in the communicated information using the existing control algorithms [18] will not be string stable over all conditions, and thus proposed the development of control algorithms that are robust to communication delays and to random dropouts. In [19], the effect of random momentary dropouts was also studied but it was for a fully redundant information structure in which all vehicles communicate their state information to each other. Similarly, in [3], measurement losses among satellites due to the shadowing effect was addressed. Full redundancy in the measurements of relative spacing was also assumed. In [20], localized estimators were proposed to reduce communication requirements. Robust stability issues were addressed whereas peak value analysis of spacing error was beyond the scope of the paper. In this paper, no redundancy is assumed. Each vehicle has only the measurements of the lead and preceding vehicles’ states as in [21]. The effect of random momentary lead vehicle state dropouts on leader-to-formation stability, as deﬁned in [22], is studied. During dropouts, the vehicles would apply the dead reckoning process to estimate the lead vehicle’s state. The resulting closed loop system is analyzed mathematically and it is determined that, for bursty links, leader-to-formation stability can be retained if the lead vehicle indeed maintains its speed and if not, the spacing error will not exceed some bound which is some ﬁnite factor of the peak value of the lead vehicle’s acceleration. II. NOTATION The notations used are as follows: refers to the -norm of the signal to the 1-norm of the im, and to the inverse transform. For pulse response a string of vehicles, we deﬁne , and as, respectively, the and distance traveled, speed and acceleration of vehicle . are, respectively, the spacing error and the desired spacing and . The symbols above a variable as between vehicles in refers to an estimate of to a propagation, to the error in the estimate and to a measurement of . III. DESIGN GOALS AND SPACING ERROR TRANSFER FUNCTION The controller design objectives are to achieve the following: • leader-to-formation stability (as in [22]) which requires spacing error to be bounded for: 1) any initial formation spacing error and; 2) any bounded input of the leader (Note

Abstract—In spacing control of multiple vehicles as in aircraft, spacecraft formation ﬂying and highway vehicle platooning, previous research has shown that the lead vehicle state is required by all following vehicles in order to achieve leader-to-formation stability. Following vehicles can estimate the lead vehicle’s state during losses by assuming that the lead vehicle maintains its speed. In this paper, it is determined that, for bursty links, leader-to-formation stability can be retained if the lead vehicle indeed maintains its speed and if not, the spacing error will not exceed some bound which is some ﬁnite factor of the peak value of the lead vehicle’s acceleration. Index Terms—Communication loss, decentralized control, leader-to-formation stability, multiple vehicles, spacing control, string stability, wireless datalink.

I. INTRODUCTION PACING control of multiple vehicles has applications in areas such as automated highway systems (AHS), aircraft [1], [2] and spacecraft [3] formation ﬂight, closely spaced parallel approaches [4], and separation control of arrival stream aircraft [5]. Spacing control research for AHS is the most extensive. There is also other related research that involves the use of the overlapping technique yielding controllers with good performance [6], and a theory and control techniques for formation maneuvering [7]–[10]. AHS research began in the 1960’s [11] and Shladover [12] has given a very extensive review of the AHS research done. Shladover [13] pointed out the need for lead vehicle information to obtain string stability with a linear static controller for spacing control. Headway control however does not require lead vehicle information [14]. Swaroop et al. [15] compared spacing and headway control laws and found that the latter required control torques inversely proportional to the headway. This could lead to input saturation for small headways. It was concluded that for high capacity, small vehicle to vehicle spacing, the constant spacing law is necessary at the price of inter-vehicle communication. The most practical means for this is through wire-

S

Manuscript received November 06, 2007; revised December 14, 2008. Manuscript received in ﬁnal form February 17, 2009. First published October 30, 2009; current version published February 24, 2010. Recommended by Associate Editor C. A. Rabbath. This work was supported in part by DSTA, Singapore, the DARPA Software Enabled Control (SEC) program administered by AFRL under Contract F33615-99-C-3014, by a DoD Multidisciplinary University Research Initiative (MURI) program administered by ONR under Grant N00014-02-1-0720; and by the DARPA MICA program administered by SPAWAR under Contract N66001-01-C-8080. R. Teo is with the DSO National Laboratories, Singapore (e-mail: rodney_teo@dso.org.sg). ´ D. M. Stipanovic is with the IESE Department and CSL, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA (e-mail: dusan@uiuc.edu). C. J. Tomlin is with the Department of Electrical Engineering and Computer Science, University of California, Berkeley, CA 94720-1770 USA, and also with the Department of Aeronautics and Astronautics, Stanford University, Stanford CA 94305 USA (e-mail: tomlin@eecs.berkeley.edu). Digital Object Identiﬁer 10.1109/TCST.2009.2016109

1063-6536/$26.00 © 2009 IEEE

For the ﬁrst vehicle behind the is aclead vehicle. . Hence would typically be in the order of 0. to achieve string stability. where is vehicle ’s vehicle as . this can be written as: . and overshoot performance speciﬁcations. . . the control input . Mathematically. When the lead vehicle state is not available. settling time. Because . let the control input be a linear combination of the accelerations. in the tually relative states dynamics and working in the -domain. The bandwidth of the vehicle system may be in the order of 1 Hz. Thus. We denote in (5) as the two components of . To study the effect of communication dropouts on spacing error while meeting the above goals. where implies a lost link between the lead implies a good link. . . where if the platoon travels mostly at constant speed. . and hicle is .. (i. . relative speeds and positions of the lead and preceding vehicles. The and th vehicle at time . we require the transfer to . and [18]. For a wireless network with broadcasts at intervals of time. . MARCH 2010 that Condition 1) is also known as string stability as in [13]). . 2. i. . refers to the lead vehicle. IV. this is a good choice where are the system matrices in (3). VOL. Substituting the equation for the input. and where (4) and . . ANALYSIS This section derives the maximum absolute value of which can be written as . and Also.1 s. NO. dynamics As mentioned earlier in [13]. and propagation is given by . • rise time. . the contribution from each input can be considered separately as if they are independent of each other. We start by writing the model of function from and . Further. V. constant) over the sample period . 18. [15]. state information of the lead vehicle is required. .470 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. . (5) (2) which implies that each vehicle assumes that the lead vehicle travels at constant speed during losses. . the control input is the same except that . a propagation of the state is computed and used in the control law. Let the relative states be and . control input and 1 to the ﬁrst vehicle behind the lead and so on. Substituting the states with the relative states in the discrete-time dynamics gives the relative states . The discrete-time dynamics of ve. . the resulting transfer function between and is (1) where . will be zero order held of vehicle at the th time interval.e.e. • satisfactorily small effect from communication dropouts on spacing error.. . The system dynamics then become a discrete-time switched linear system (3) where the state and control inputs are . HANDLING COMMUNICATION DROP-OUTS The method studied here to handle communication dropouts is to control with a propagation of the lead vehicle state.

. In fact. The second conclusion can be stated as the following theorem. these results can be used as a design guide that gives the relationships among the peak acceleration permitted. Outline of proof: The proof can be derived by individually examining the terms of the expansion of the second term of (5). the worst-case effects of . The following can thus be written: (6) . impulse response of (1). First.TEO et al. and or when . To choose the control law gains. Therefore where To ﬁnd that maximizes . then . we ﬁnd those that minimize the effect of datalink losses on . we can consider the effects of (8) with and . One of the vehicle’s link has to be UP for there to be an impact on spacing control. would be given by the application of the worst-case link loss and switching sequence [s]. can be computed. will be attenuated over time. NUMERICAL EXAMPLE The case of 3 s. and . could grow. These numbers represent typical systems which would have speed control settling times of around 3 seconds and wireless links enabling 10 Hz small packet broadcasts with losses not lasting more than 1 s. the system retains string stability even in the presence of link losses in that upstream spacing error disturbances attenuate as they propagate downstream along the string. this is because the lead vehicle is predicted to travel at constant speed during link losses. Theorem 1: Let the settling time of the spacing controller under no link loss conditions be and the value of be computed for a link outage duration of . It can be seen by inspection of (5) that the products involving (for all ) are independent of the link status parameter. Intuitively. there is an intuitive explanation for this. This nulliﬁes the effect of link losses on this component of the transfer function. On hindsight. Consider instead bursty links which will be UP for most of the time and DOWN sporadically. they are in fact computing their controls with respect to a virtual but common vehicle. in the presence of lead vehicle accelerations and upstream disturbances in spacing errors. in the absence of lead vehicle accelerations and with the appropriate choice of controller gains. (Proposition 1) for the period This shows that the spacing error of all vehicles in a string will be bounded by some ﬁnite factor of the peak value of the acceleration of the lead vehicle. For small samquences that maximize the gain from to . . the transfer function to is identical to the case in which the links are from always UP. and pling period which implies [see (1)]. (10) (7) Remark 1: It implies that the worst case condition occurs when either the vehicle’s link or its preceding vehicle’s link is down BUT not when both are down simultaneously. If we assume that the settling time of system is less than or equal to [s]. it is guaranteed that the spacing error will be bounded as in (9). the spacing error can be bounded as per (8). and monotonically decreases to if . Last. are cascaded together leading to a worstvehicle case accumulative effect on vehicle as in (9) . . the link outages are less than peak acceleration is and the outages are spaced out by at least . Let [s] be the minimum time the links remain UP in between outages. where is the Thus. [refer to Proposition 1: If for and if . there is no impact on spacing control. Using (8). maximum absolute value of is obtained when . is unbounded with It can be seen from (7) that increasing . Thus. 1 s is now studied to check the analysis above. the gradient. For can be seen that monotonically increases up to . . when both links are DOWN. For real-life implementation. letting and both . the link loss statistics and the peak spacing error that can be tolerated. If the lead vehicle’s . and with link losses. it if boundedness attenuation. The expansion gives Now. . then the (1)]. This may not usually occur in practice but would give a guaranteed upper bound on . When they are restored DOWN for to UP. where is the induced -norm from to Consider now the contribution from as represented by which can be obtained by ﬁnding the switching and input se. This part of the system is in fact equivalent to an LTI system. is derived. and hence as well. If both and have lost links with the lead vehicle and are vehicles estimating the lead vehicle state simultaneously. VI. and for . . The analysis above can now be summed up as follows. When they are [s].: DECENTRALIZED SPACING CONTROL OF A STRING OF MULTIPLE VEHICLES OVER LOSSY DATALINKS 471 Consider ﬁrst the contribution from .

With reference to (10). NO. MARCH 2010 Fig. .. The minimization problem produces the gains. over subject to Problem: To minimize (rise time is set to less than ). the spacing error will not exceed some bound which is some ﬁnite factor of the peak value of the lead vehicle’s acceleration. A steepest descent gradient search is used together with penalty functions. those that minimize which is computed for an . is shown in Fig. cient is set to greater than or equal to . this also sets a limit ( is chosen to be less than on the overshoot). (settling (damping coefﬁtime is set to less than ). Future work might be to estimate the lead vehicle’s state by assuming that the last known acceleration is maintained. the vehicles estimate the lead vehicle’s state during losses by assuming that it would maintain its speed. a Monte Carlo simulation of 500 runs is performed. loop system of (5) be Recall that -domain and -domain poles with identical time. VII. The results. and damping coefﬁcient in the ﬁrst three constraints are good for second order systems [23]. shows that for ). are in the unit circle for system stability) and (control gains are bounded which will in turn limit the size of the control inputs). and values: 0. Note that the spacing error attenuates from vehicles 1 to 4 and thereafter are clearly bounded. This . leader-to-formation stability can be retained if the lead vehicle indeed maintains its speed and if not. . .472 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. VOL. Time history of 10 vehicles following lead vehicle’s 4-s 1 m/s acceleration. Let the dominant poles of the closedoutage duration of and let . The design is performed with the following . i. for bursty links. as in (as computed Table I. 1. the simulation for 10 vehicles led by a lead vehicle which performs an acceleration of 1 m/s for 4 s causing the speed of the string to increase from 20 to 24 m/s. 18. For the study. The acceleration input of all vehicles appear noisy. The following controller design problem is thus proposed. with . 2. 1 to achieve string stability during perfect communications). where is the domain characteristics are related by sampling period. It is determined that.e. (ks k )=(ka k ) WITH q = 0:6. settling time. This is due to the changing link status. 1. CONCLUSIONS AND FUTURE WORK This paper studies the effect of intermittent losses of the lead vehicle’s state on leader-to-formation stability. Emulating the links as ﬁrst order Markov models where the probability of the link going UP when previously UP is and when previously DOWN is . The approximations of rise time. this corresponds to a case of “boundedness”. (system poles.1 s. which give . r = 0:6 FOR VARIOUS CUMULATIVE PROBABILITIES TABLE I To check that the spacing error bound is respected.

and C. 1520–1533. vol.. 434–442. Technol. Dyn. Control. Shladover. pp. C. 657–672. Tomlin. Pachter. Control. May 2002. Ogren. Speyer. 2463–2470. and C. Autom. 2003. 401–420. Barcelona. vol. F. G. pp. no. 8. 4. and C. Olfati-Saber. 2836–2841.” IEEE Trans. 1995. Control. Guid. [23] G. D.” Tech. Lavaei. J. 2001. 9. Singapore. vol. [13] S. D’Azzo. [21] R. Chien. Autom. pp.” Veh. 6–7. vol. “Computing danger zones for provably safe closely spaced parallel approaches. 3. and A. D.. 42. Control.” IEEE Trans. 3. [9] R. 49. 1999. 2004.. no. 2004. Control Conf. G. Dec. Pappas. “Tight formation ﬂight control. 3107–3112. F. J. Mar.” Automatica. vol. [20] J. Control Conf. Autom. Desoer.: DECENTRALIZED SPACING CONTROL OF A STRING OF MULTIPLE VEHICLES OVER LOSSY DATALINKS 473 may be a more accurate estimate in cases where the links recover faster than lead vehicle acceleration changes.” IEEE Trans. and K. [22] H. Mahal. D. Park. “Longitudinal control of a platoon of vehicles with no communication of lead vehicle information: A system level study. loannou and C. VA.” ASME J. G. CO. Technol. 2006. “Peak-seeking control with application to formation ﬂight. 8. Tomlin. Hanson. 3. pp.” IEEE Trans. “Information ﬂow and cooperative control of vehicle formations. [12] S. Momeni. Control. J. Aug.. [8] R.. Jun. pp. A. no. M. Project METRAN: An Integrated. 1998. 433–455. no. “Decentralized overlapping control of a formation of unmanned aerial vehicles. Hedrick. “Leader-to-follower stability. 42.” presented at the 4th Asian Control Conf. loannou. Spain. pp. Vinken. 546–54. CA: Addison-Wesley. [14] P. Control. vol. 1993.” in Proc. [15] D. and C. vol.” Veh. J. Kumar. Syst. M. no. E. [5] P.” in Proc. Control Conf. Dyn. ´ [6] D. and S. Fiorelli. “A comparison of spacing and headway control strategy for automatically controlled vehicles. L. 24. 1966. 40. 2002. Stipanovic. 2003. 26. Inalhan. pp. Control Syst. Leonard. pp. 2. [3] R. G. UCB-ITS-PRR-2001-29. pp. Chen. “Flocking for multi-agent dynamic systems: Algorithms and theory. A. no. pp. Digital Control of Dynamic Systems. Y. MA: MIT Press. pp. Rep. Nov. 20. 4. Sheikholeslam and C. Hedrick. McMahon. Syst. Sep. Nav. 682–687... Tomlin. “Longitudinal control of automated guideway transit vehicles within platoons. Robot. 1285–1296.” presented at the 15th IFAC World Congr. Veh. Shladover. pp. Tomlin. Workman.” AIAA J. Stipanovic. California Partners for Advanced Transit and Highways. Hadaegh. 2004. E. 246–254. Dyn. [16] S. M. and V. Jul. vol. IEEE Conf. . E. Oct. Smith and F. no. and P. 1993.. no. no. Syst. REFERENCES [1] D. [18] J. vol. A. Narendran. 23. 49. Guid. 1991. and M. J. Teo. and A. Y. pp. “A model predictive decentralized control scheme with reduced communication requirement for spacecraft formation.. vol. Amer. J.” in Proc.TEO et al. S. AIAA Guid. Technol. Murray. Teo. pp. “Optimized vehicle control/communication interaction in an automated highway system. “Cooperative control of mobile sensor networks: Adaptive gradient climbing in distributed environment.. vol. [17] J. 2002. Swaroop. Control. 1292–1302./Apr. Franklin. J. “Control topologies for deep space formation ﬂying spacecraft. Evolutionary Transportation System for Urban Areas. G. Fax and R. J.. no. Aug. [7] J.” AIAA J. J.. Veh. “Consensus problems in networks of agents with switching topology and time-delays. E. 1994.. Autom. pp. Proud. “Review of the state of development of advanced vehicle control systems. 51. Mar. Hoffman. 8. Amer. IEEE CDC. 268–278. [2] M. W. Arlington. J. Nav. AIAA2000–AIAA-4362. V. Tanner. 24. L. Meas. Nov.” IEEE Trans. 2001.” in Proc.” IEEE Trans. and D. Chichka. pp.. 100. 3. Jul. Mar. Stipanovic. Teo and C. 4. C. Murray. vol. pp. Menlo Park. Nov. Nav. [11] M. R. 2. Teo. 1978. Sep. Swaroop. “Inﬂuence of speed and altitude proﬁle on the dynamics of in-trail following aircraft. Powell.” in Proc. “Multiple vehicle control over a lossy datalink. 2000. no. Jun. Aghdam. “Intelligent cruise control. 16. [4] R. “Longitudinal vehicle controller design for IVHS systems. 597–625. Chien. no. 291–297.. vol. “Decentralized spacing control of a string of multiple vehicles over lossy datalink. no. D. Denver. ´ [19] R.. Olfati-Saber and R. [10] P. 2008. pp. Cambridge. M. Hedrick. J. Zeghal.” IEEE Trans. J. vol. 551–595. May/Jun. 2004. and N. Dec. M..