adaptive communication scheme for cooperative

ADAPTIVE COMMUNICATION SCHEME FOR COOPERATIVE ACTIVE SAFETY SYSTEM Shahram Rezaei1, Raja Sengupta2, Hariharan Krishnan3, Xu Guan4 1

PhD, Senior Researcher California PATH, UC Berkeley 1357 S.46th Street, Richmond, CA 94804 USA +1-510-665-3552, [email protected]

2

Associate Professor Civil Eng. Dept., UC Berkeley 707 Davis Hall, UC Berkeley CA 94720 +1-510-665-3552, [email protected]

3

PhD, Staff Researcher Electrical & Controls Integration Laboratory GM R & D Center 30500 Mound Road, Warren, MI 48090 +1-586-986-6966, [email protected]

4

PhD Student Civil Eng. Dept., UC Berkeley 604 Davis Hall, UC Berkeley CA 94720 +1-510-642-9569, [email protected]

Abstract- We present an adaptive communication scheme for Cooperative Active Safety System (CASS). CASS uses information communicated from neighboring vehicles via wireless communication in order to actively evaluate driving situations and provide warnings or other forms of assistance to drivers. In CASS, we assume that vehicles are equipped with a GPS receiver, a Dedicated Short Range Communications (DSRC) transceiver, and in-vehicle sensors. The information exchanges between vehicles include position, speed, heading, and other vehicle kinematic and dynamic information, and the information is broadcast to all neighbors within a certain communication range. The literature surmises CASS may need a vehicle to broadcast information as often as every 100 msec which may lead to channel congestion resulting in message loss rates above 20%. Here we present a new communication design scheme, supported by simulations, which indicates that CASS could be enabled by broadcasting, on average, as little as once every 500 msec. Key Words- V2V, Safety, DSRC, Communication INTRODUCTION In this paper, we describe a new communication scheme for Cooperative Active Safety System (CASS). An Active Safety System is an in-vehicle system that provides warnings or other forms of assistance to drivers based on information obtained from in-vehicle object detection sensors about the motions of other vehicles in its field of view (1)-(2). Active safety systems rely on object detection and ranging sensors such as radars, lidars, etc. in order to monitor vehicles in the field of view. CASS aims at accomplishing this goal by using technology alternatives such as GPS and Dedicated Short Range Communications (DSRC) and also has the potential to provide information from vehicles that may be occluded from direct line of sight. In CASS, we assume that vehicles are equipped with a GPS receiver, a DSRC transceiver, and in-vehicle sensors. The information exchanged between vehicles consists of position, speed, heading, and other vehicle kinematic and dynamic information, and this is broadcast to all neighbors within a certain communication range (or power). Applications aboard the

-1-

CASS vehicles receive this information and actively evaluate driving situations and provide warnings or other forms of assistance to drivers. CASS is now a well established concept in the literature (3)-(11). We built the first Wi-Fi and GPS based prototype able to provide warnings for a variety of driving situations on sub-second time-scales (12). Our design also resolved some of the challenges on reliably estimating vehicle position (13). This paper presents a new communication design intended to be used by CASS vehicles to track each other as required, and to provide driver assistance. Prior communications designs for these applications have relied on periodic broadcasts by each vehicle that are sometimes supplemented by event-driven messages. For example, a vehicle might broadcast motion information such as position, speed, heading and so on, at regular intervals. It may also occasionally broadcast an extra message triggered by an event such as emergency hard braking. By contrast, the communication design explored here is adaptive in nature. The Vehicle Safety Communications Consortium (VSCC), formed by the U.S. automotive industry under cooperative agreement with the USDOT conducted a project to investigate cooperative Vehicle Safety Communications (VSC) applications and technology. The VSC project extensively investigated the periodic and periodic plus event communication model. The recommended communication rate for all but two applications is a message every 100 msec per vehicle (14). By contrast, the communication design presented here is designed to be adaptive and may be able to support VSC applications with a message every 500 msec on average per vehicle. This potential saving in the broadcast rate is the motivation for our work. At a fixed message broadcast rate every 100 msec per vehicle, it is expected that the DSRC channel congestion would be severe resulting in message loss probabilities that may be over 20%. At an average message broadcast rate every 500 msec per vehicle, the DSRC channel would be relatively less congested resulting in message loss probabilities that may be between 4 and 5%. Several researchers including us have quantified message loss rates or delays for the periodic or periodic plus event message designs (15)-(18). The performance numbers obtained depend on assumptions about the channel, the physical or medium access control (MAC) designs, message size, transmission power, modulation, and coding. Therefore the emphasis of this paper is more on establishing the inter-vehicle messaging rate for vehicle communications rather than the communication performance that will result from the rate. The paper is organized as follows. We describe the design first. Next, we describe the simulation model used to estimate the message rates resulting from the design. The performance results are described next.

DESIGN Fig. 1 shows the design block diagram that we propose for CASS. Each vehicle in CASS has an extended Kalman filter, called the “Self Estimator” that estimates its position, speed, and heading by integrating differential GPS and the vehicle sensors (including speed, steering angle, and yaw rate sensors). These quantities collectively constitute the vector Xˆ (k ) in Fig. 1. Xˆ (k ) is the best available estimate state of the vehicle at time k. Details of the Self Estimator can be found in (13). For the rest of the discussion, we refer to Xˆ (k ) of the SelfEstimator as the state of the vehicle at time k.

-2-

Fig. 1. Design Block Diagram We also require each vehicle to run an estimator for each vehicle in its neighborhood. This is called the “Neighbor Estimator” in Fig. 1. This estimator receives the messages reporting the state Xˆ (k ) of the neighboring vehicle. In the figure, OV (Other Vehicle) receives the messages of SV (Subject Vehicle). OV runs a model to provide its vehicle with an estimate of the state of the SV for times in-between message receptions from SV. The outputs of the vehicle Self Estimator and the Neighbor Estimators for all the neighboring vehicles should drive the responding applications in the vehicle. If a vehicle (say SV) has n neighbors (OV’s) it will have to run n Neighbor Estimators. There is a third estimator, called the “Remote Estimator” in Fig. 1. This exists solely to enable our communication design. Each vehicle has one Remote Estimator. Let the Remote Estimator of vehicle i be denoted by REi. Let the Neighbor Estimator run by neighbor j for vehicle i be denoted NEji. The purpose of REi is to estimate the output of all the NEji’s. It is an estimator of all the Neighbor Estimators. A vehicle’s decision to communicate or not communicate at any instant of time is taken based on the difference between the outputs of its Self Estimator and Remote Estimator. The inputs to its Remote Estimator are the messages broadcast by the vehicle to its neighbors. The Remote Estimator outputs estimates what the neighbors are thinking about the vehicle, i.e., the neighbor’s estimates of Xˆ (k ) . If the difference between the Xˆ (k ) and the output of the ~ Remote Estimator X (k ) exceed a threshold at any time k, it results in a broadcast of Xˆ (k ) by

the vehicle. More precisely, the scheduler in Fig. 1 looks at the differences between the self and the Remote Estimator’s estimates and decides to broadcast or not, according on the following policy: u (k ) = 1 if (ε long . (k + 1) > Tr.long . ∨ ε lat . (k + 1) > Tr.lat . ) u (k ) = 0 Otherwise

where, k, u(k),

is the time index, is the scheduler’s decision on communication at time k (1 means communication and 0 means no

-3-

(1)

communication) !long.(.), is longitudinal tracking position error, !lat.(.), is lateral tracking position error, Tr.long. is the threshold on the longitudinal error, is the threshold on the lateral error. Tr.lat.

~! !ˆ The state vectors X (.) , X (.) consist of positions X, Y (in the universal GPS coordinate frame with a local origin), speed V, and heading angle φ. Symbols with (^) on top refer to the Self Estimator’s estimates and symbols with (~) on top refer to the Remote Estimator’s estimates. In this paper, all communications are triggered by thresholds on differences in the longitudinal and lateral positions output by the Self and Remote Estimators. We calculate the longitudinal and lateral position differences between estimates of the Self and the Remote Estimators using equation (2). Longitudinal and lateral errors are position errors along and normal to the SV’s direction of motion.

( ) (~ ) ( ) (k ) = (X~ (k ) − Xˆ (k ))× sin (φˆ(k ))+ (Y~(k ) − Yˆ (k ))× cos (φˆ(k )) ; (~

)

ε long . (k ) = X (k ) − Xˆ (k ) × cos φˆ(k ) − Y (k ) − Yˆ (k ) × sin φˆ(k ) ; ε lat .

(2)

The Remote and Neighbor Estimators use the same model. We keep the model simple because a vehicle has to run many Neighbor Estimators, i.e., one for each neighbor. The Remote Estimator estimates by running the following Kinematic equations.

~ ~ ~ ~ X (k + 1) = X (k ) + V (k )× cos φ (k ) × ∆T ; ~ ~ ~ ~ Y (k + 1) = Y (k ) + V (k )× sin φ (k ) × ∆T ; ~ ~ V (k + 1) = V (k ); ~ ~ ~ φ (k + 1) = φ (k ) + φ" (k ) × ∆T ; ~ ~ φ" (k + 1) = φ" (k );

(

(

)

)

(3)

The state estimator of the Remote Estimator is reset to that of the Self Estimator whenever the SV broadcasts a message. Correspondingly, that of a Neighbor Estimator is reset to the state in a received message whenever one arrives from the OV being tracked. Shladover and Tan (19) have derived the accuracies required in position, speed, and heading estimates to produce warnings of reasonable accuracy and consistency. We use their work to set the thresholds for the evaluations. The lateral position errors requirements turn out to be more stringent than the longitudinal error requirement. If the communication channel has no loss or delay, the Remote Estimator estimates exactly what each neighboring vehicle estimates as the state of vehicle i. This is because each time the vehicle sends a message reporting Xˆ (k ) to its neighbors, it also sends it to its own Remote Estimator as show in Fig. 1. We also have the Remote Estimator run the same model as that run by the Neighbor Estimators. Thus identical equations driven by identical inputs will produce identical outputs. Moreover the difference between the state of the vehicle at

-4-

time k and what a neighbor thinks of as the state of the vehicle at time k is always less than the threshold used to trigger the next broadcast of Xˆ (k ) . Assuming the threshold is chosen to be small enough to be tolerated by the responding applications, the design would provide applications in CASS with a sufficient estimate of the state of the sender. All these arguments hold exactly when there is no communication loss or delay. Since messages will actually be lost, the difference between Xˆ (k ) and the neighbors estimate of Xˆ (k ) will sometimes exceed the threshold used to trigger messages. This threshold violation may result in violation of the position accuracy required by some applications. We think of this as a cost. If the communication loss probabilities are small, these costs will be small and perhaps acceptable. We are working on enhancing the adaptive communication scheme to be robust to communication losses. Communication delays also cause threshold violations. If the difference between the Self Estimator and Remote Estimator outputs crosses the threshold at time t, it will trigger a broadcast of Xˆ (t ) that may be received at time t+δ by a neighbor. During the interval δ the error in the neighbor’s estimate will exceed the requirement. For example if δ is 50 msec, the accuracy requirement will be exceeded by 0.024m at most if we assume the maximum deceleration of a vehicle is limited to 1g. A threshold violation by 0.024 meter could be insignificant. GPS errors themselves are likely to be much larger. Communication loss is the more significant concern.

EVALUATION METHOD The rate at which messages are produced by our design will depend on the dynamics of vehicles. If the state of a vehicle is not changing much, the kinematic equations run by neighbors will accurately predict the state of the vehicle. The differences between Self and Remote Estimator outputs will remain small and few messages will be produced. On the other hand, fast changing dynamics will result in more frequent messages. Thus to evaluate the communication rates produced by this design we have incorporated our design into a vehicular traffic simulator. We have then simulated different kinds of traffic to understand which scenarios might generate the highest rates of communication. In this section we describe the simulation tool. We use the Ruby SHIFT (20) and Paramics (21) traffic simulators. SHIFT has been developed by California PATH at University of California at Berkeley. Fig. 2 is snapshot of a typical simulated section. The maximum per lane traffic flow of our vehicle traffic simulator is around 2250 vph (vehicles per hour) per lane when desired inter-vehicle gap is 1.1 sec. At this flow, the average inter-vehicle distance is about 30 m, density is 55 vpm (vehicles per mile) per lane, and the average speed is 27 m/s.

Fig. 2. Snapshot of ruby SHIFT traffic simulator

-5-

Inputs to the simulator are roadway parameters such as the length of sections, number of lanes, desired vehicle inflows, speeds, and inter-vehicle gaps. Vehicles can make lane changes. The output of the simulator is a log file of positions X, Y, and speed (V) of vehicles at 20 Hz. We assume these are the true values of these quantities. We consider the heading H to be zero when the vehicle is not changing lanes. Vehicles change lanes instantaneously in the traffic simulator which is not realistic. Therefore, we adjust the vehicle’s trajectory for the duration of the lane change in order to make it smooth during the lane change period. At each lane change, we pick a random number (uniform distribution) between 2 and 3 seconds to be the lane change duration. Fig. 3 shows how the design is modeled as a block diagram.

Fig. 3. Rate control scheme implementation using traffic simulator data Based on the true position, heading, and speed of vehicles output by the simulator we model the Self and the Remote Estimators’ estimates as follows: 1) We add Gaussian noise to X, Y, V, H, and the heading rate ( H" ) in order to simulate errors in the output of the Self Estimator. Standard deviations of noise are 0.2 m, 0.2 m, 0.2 m/s, 1 degree, and 0.3 deg/s, respectively. These numbers match experimental data (12). The additive noises are correlated by a first order autoregressive (AR) model. This is because our experimental data shows the Self Estimation errors are colored rather than white. We use the same AR model for the four states. The model for the noise denoted w is: w(k + 1) = α w(k ) + β z (k ) ;

α = 0.9 ; β = 0.436 ; z (k ) = N (0 , σ w )

(4)

where, z(k) is zero mean white Gaussian noise with standard deviation equal to the standard deviation of ‘w’ in steady state. 2) Calculate the Remote Estimator’s estimation of states using equations (3). 3) Calculate the longitudinal and the lateral tracking errors using equations (1) and generate a message if at least one of the errors exceeds the related threshold. 4) Log all the time traces of the messages produced by all the vehicles and use them to produce the results in section 0.

-6-

RESULTS In this section we present two sets of results. We quantify the communication load generated by our design. We present simulation results for three vehicular traffic scenarios: regular freeway, freeway with a merge lane, intersection. We use the Ruby SHIFT simulator for the first two and the Paramics simulator for the intersection scenario. We use 0.5 m and 0.3 m as values for the longitudinal and lateral errors thresholds, respectively, unless stated otherwise. Using smaller threshold values trivially results in faster communication, while larger threshold values require less communication. We conservatively chose 0.3 m for the lateral error threshold because warning applications require the standard deviation of the lateral error to be smaller than 0.5 m (19). Our performance measure is the number of messages per second per meter. To avoid fractions we calculate the number of message per 300 meters per 2 seconds at every point along the roadway. This is because the capacity of a wireless network is better understood in bit-meters/second. The bit per second per vehicle measure does not capture the workload created by increasing communication range.

Regular Freeway Fig. 4 presents the spatial density of communication for a straight section of a freeway with our design. Specifications of this test are: Road length: 6 km, Sampling rate: 20 Hz, Average flow: 1900 vph per lane, Average inter-vehicle gap: 1.1 sec Number of lanes: 4, Longitudinal error threshold: 0.5 m, Lateral error threshold: 0.3 m For this run, the average inter-vehicle distance is 40 m. Hence, the density is 40 meters per vehicle per lane. At maximum flow density is 29.3 meters per vehicle per lane (55 vehicles per mile per lane). Thus, the traffic is flowing freely. If periodic communication were happening at 10 Hz (i.e. without rate control), within a 300m distance window and 2 sec time window, the number of message produced per 300 meters and per 2 seconds would be 600 as below: 300 [m] 600 = × 4 [lanes ]× 2 [sec]×10 & samples # sec!" $% 40 [m]

Fig. 4. Communication density in a regular 4 lane freeway

-7-

Fig. 4 shows the same number with this design at all points on the freeway is less than 100 messages. This reduces the communication load by 1/6th. Fig. 5 shows how the average communication message time interval per vehicle is changing at different traffic flow conditions. The flow values cover both free flow and jammed traffic conditions. When entering the jammed region, flow is decreasing while density is increasing. The arrows indicate, the traffic condition switching from freely flowing to jammed condition.

Fig. 5. Variation of the average communication message time interval versus traffic flow Communication is triggered when either the longitudinal or the lateral error exceeds the threshold. Fig. 5 shows, there is small variation in the average time interval between messages in the free flow traffic region. Its mean is at 0.520 sec and its standard deviation is just 0.0267 sec. This is because at higher flow there is more net acceleration (i.e. average absolute value of acceleration per vehicle) which results in faster growth of the longitudinal error, triggering more crossings of the longitudinal error threshold as indicated by the red line. On the contrary, in the freely flowing region, as flow increases there is less space available for lane changes. This reduces the number or lane changes. Speed also reduces at higher flows. The two reductions minimize the number of lateral threshold crossings as indicated by the blue line. The increasing rate of longitudinal threshold crossing and reduced rate of lateral threshold crossings counter balance each other to result in an average time interval between messages of about 500 msec in the free flow region regardless of traffic flow. Fig. 6 shows how speed and acceleration are changing with flow. One can see the maximum acceleration conditions occur as the traffic transitions from freely flowing to jammed. This shows the transition point should represent the highest rate of longitudinal threshold crossing.

Fig. 6. Variation of average speed and net acceleration versus flow

-8-

In the jammed case, even less communication is required because there is very little net acceleration in the system, speed is small, and lane changes are rare. Thus, jammed traffic is not the highest communication scenario. Choosing smaller and larger threshold values results in higher and less communication, respectively. The relationship between threshold value and the average time interval between messages is almost linear. Fig. 7 shows the relationship. In the figure, threshold factor is the ratio of the threshold values used to determine the average message time interval and the nominal values. The nominal values are 0.5 m and 0.3m for the longitudinal and the lateral error threshold. The same threshold factor is used for both the errors in Fig. 7.

Fig. 7. Variation of the average communication message interval versus threshold value factor

Merge Fig. 8 presents results for a 3 lane freeway with an added merge lane. The merge lane joins the freeway at 1000 m. Specifications of this test are: Road length: 7 km, Sampling rate: 20 Hz, Average flow: 1700 vph per lane Average inter-vehicle gap: 1.1 sec Number of lanes: 3 plus one merge lane Longitudinal error threshold: 0.5 m, Lateral error threshold: 0.3 m

Fig. 8. Communication density in a freeway with a merge lane -9-

The average flow is 1700 vph per lane across four lanes. When lane four merges with lane three the flow becomes 1700*4/3 = 2267 vph per lane. This is the maximum flow condition. Hence we have simulated this scenario. Moreover, merges may result in larger numbers of lane changes. For this run, the average inter-vehicle distance is 37 m near the merge zone (i.e. around 1000m). If the communication was happening at 10 Hz (i.e. without our design), within the 300m distance window and 2 sec time window of Fig. 8, on average 645 messages would be produced as calculated below: 300 [m] 645 = × 4 [lanes ]× 2 [sec]×10 & samples # sec!" $% 37 [m] Using the design we see even at the merge zone less than 160 messages are generated.

Intersection Finally, we simulate a suburban intersection. We do this for two reasons. It has been argued that the braking and acceleration of vehicles caused by the change of the traffic light may result in many messages. Secondly, suburban roads can carry a lot of traffic. Specifications of this test are: Road length: 1 km, Sampling rate: 20 Hz, Average flow: 1700 vph per lane, Number of lanes: 4 Number of approaches: 4, Number of departures: 4 Traffic light duration: 50 sec (20 sec green, 20 sec red, 2 sec all read, 3 sec yellow) Speed limit: 50 km/hr Longitudinal error threshold: 0.5 m, Lateral error threshold: 0.3 m Fig. 9 shows the average communication density for the departing directions. Distance if measured from the intersection. For this run, the average inter-vehicle distance is 10 m. If periodic communication were at 10 Hz (i.e. without this design), within a 100m distance window and 2 sec time window used in Fig. 9, on average 800 messages would be produced as calculated below: 100 [m] 800 = × 4 [lanes ]× 2 [sec]×10 & samples # sec!" $% 10 [m]

Fig. 9. Average communication density along the directions departing the intersection -10-

By contrast, Fig. 9 shows the number of message using the design is less than 60. The highest number of messages occurs right near the intersection. Thus the starting and stopping causes communications as one might expect. The communication rate dips as the vehicles move away and rises again. The rise is most likely associated with the increase in speed as also observed in our regular freeway simulation. Fig. 10 shows the number of messages for the approaching directions.

Fig. 10. Average communication density along the directions approaching the intersection

CONCLUSION In this paper, we designed an adaptive communication scheme in order to lower the intervehicle communication rate in CASS. The adaptive communication scheme is designed to bound the longitudinal and lateral position tracking errors relative to other communicating vehicles within predefined bounds required for the application functionality. In the proposed scheme, each vehicle compares its self state estimates of the state outputs of their remote estimator. Then based on a threshold crossing policy defined for longitudinal position error and lateral position error violation, the vehicle decides to broadcast or not. Using traffic simulators, we have evaluated the performance of the proposed scheme using a number of simulations based on a variety of driving environments that includes regular highway, merge zone, and intersection. Although the literature surmises that CASS would need a vehicle to broadcast information as often as every 100 msec, we have presented a new communication design scheme, supported by simulations, which indicates that CASS could be enabled by broadcasting, on average, as little as once every 500 msec.

ACKNOWLEDGMENT

We are very thankful to Mr. Joel VanderWerf in California PATH who provided the SHIFT based traffic simulator to us. Research supported in part by General Motors R&D Center through contract #TCS 70709 to UC Berkeley.

-11-

REFERENCES

(1) NHTSA, ITS Joint Program Office,"Report to Congress on the National Highway Traffic Safety Administration ITS Program, Program Progress During 1992-1996 and Strategic Plan for 1997-2002". U.S. Department of Transportation, Washington, DC, January, 1997. (2) D. N. Godbole, R. Sengupta, J. Misener, et. al., “Benefit Evaluation of Crash Avoidance Systems”. Transportation Research Record No: 1621, 1998. pp 1-9, January 1998. (3) F. Broqua, G. Lerner, V. Mauro and E. Morello, “Cooperative Driving: Basic Concepts and a First Assessment of ‘Intelligent Cruise Control’ Strategies”, Proceedings of Advanced Telematics in Road Transport, DRIVE Conference, Vol. II, pp. 908-929, February 1991. (4) E. Schubert, L. Reck and J. Graf, “Communication Systems for Cooperative Foresighted Driving”, Proceedings of Fourth Annual Meeting of ITS America, pp. 374–381, 1994. (5) H. J. Asher and B.A. Galler, “Collision Warning Using Neighboring Vehicle Information”, Proceedings of Sixth Annual Meeting of ITS America, pp. 674–684, 1996. (6) H.J. Asher and B.A. Galler, “Collision Warning in a Mix of Equipped and Unequipped Vehicles”, Proceedings of Fourth World Congress on Intelligent Transport Systems, 1997. (7) Z. Yang, T. Kobayashi and T. Katayama, “Development of an Intersection Collision Warning System Using DGPS”, SAE Paper No 2000-01-1301, in Intelligent Vehicle Systems, SAE Special Publication SP-1538, pp. 123-127, 2000. (8) C. Passmann, C. Brenzel and R. Meschenmoser, “Wireless Vehicle to Vehicle Warning System”, SAE Paper No 2000-01-1307, in Intelligent Vehicle Systems, SAE Special Publication SP-1538, pp. 149160, 2000. (9) A.A. Oloufa and A.E. Radwan, “Server-Based Collision Avoidance Using DGPS”, Proceedings of Eighth World Congress on Intelligent Transport Systems, 2001. (10) S. Kato, S. Tsugawa, K. Tokuda, T. Matsui and H. Fujii, “Vehicle Control Algorithms for Cooperative Driving with Automated Vehicles and Intervehicle Communications”, IEEE Transactions on Intelligent Transportation Systems, Vol. 3, No. 3, pp. 155-161, 2002. (11) S. Muller, M. Uchanski and J.K. Hedrick, “Cooperative Estimation Using Vehicle Communication”. Vehicle System Dynamics, Vol. 39 No. 2, pp. 121-133, 2003. (12) R. Sengupta, S. Rezaei, S. E. Shladover, J. A. Misener, S. Dickey, H. Krishnan, “Cooperative Collision Warning Systems: Concept Definition and Experimental Implementation”, Journal of Intelligent Transportation Systems, in review, 2006. (13) S. Rezaei, R. Sengupta, “Kalman Filter Based Integration of DGPS and Vehicle Sensors for Localization”, IEEE Transactions on Control Systems Technology, Accepted for publication, 2006. (14) Vehicle Safety Communications Consortium (VSCC), “Vehicle Safety Communications Project: Task 3 Final Report: Identify Intelligent Vehicle Safety Applications Enabled by DSRC”, 2005. (15) J. Yin, T. ElBatt, G. Yeung, B. Ryu, S. Habermas, H. Krishnan, T. Talty, “Performance Evaluation of Safety Applications over DSRC Vehicular and Ad Hoc networks“, In Proc. of the 1st ACM Workshop on Vehicular Ad Hoc Networks (VANET 2004), 2004. (16) Q. Xu, T. Mak, J. Ko, and R. Sengupta, “Vehicle-to-Vehicle Safety Messaging in DSRC”, In Proc. of the 1st ACM Workshop on Vehicular Ad Hoc Networks, pp. 19-28, 2004. (17) M. Torrent-Moreno, D. Jiang, H.s Hartenstein. ”Broadcast Reception Rates and Effects of Priority Access in 802.11-based Vehicular Ad-Hoc Network”, In Proc. of the 1st ACM International Workshop on Vehicular Ad Hoc Networks (VANET 2004), 2004. (18) T. ElBatt, S. Goel, G. Holland, H. Krishnan, J. Parikh, ”‘Cooperative Collision Warning Using Dedicated Short Range Wireless Communications”’, In Proc. Of the 3rd ACM International Workshop on Vehicular Ad Hoc Networks (VANET 2006), 2006. (19) S. E. Shladover, S.-K. Tan, “Analysis of Vehicle Positioning Accuracy Requirements for Communication – Based Cooperative Collision Warning”, Journal of Intelligent Transportation Systems, in review, 2005. (20) SHIFT Homepage: http://path.berkeley.edu/SHIFT/ (21) PARAMICS Homepage: http://www.paramics.com/

-12-