Securing Wireless Communications of Connected Vehicleswith Artificial Intelligence Review Paper ABSTRACT: This paper studies theapplication of artificial intelligence (AI) in the upcoming field of autonomousvehicles. Self-driving cars need seamless, accurate exchange of position andsafety information for navigation, which is provided by the Vehicular Ad-HocNetwork (VANET). Following the model of a learning-based agent, the AI systemcontinuously learns to augment its ability in discerning and recognizing itssurroundings. VANET is susceptible to threats such as disabled braking, falsemeter readings and unauthorized controls by spoofed messages being insertedinto the network from external sources.
However, as of today the standards leadto an offset in the limited processing ability of the security measures versusthe validation of large number of messages being circulated in the VANET,making it subject to denial-of-service (DoS) attacks. This interdisciplinaryresearch shows promising results to reach an acceptable trade-off betweenmessage authentication and DoS prevention. Message authentication adoptsContext-Adaptive Signature Verification strategy, applying AI filters to reduceboth communication and computation overhead. Combining OMNET++, a data networksimulator, and SUMO, a road traffic simulator, with Veins, an open sourceframework for VANET simulation, the study evaluates AI filters comparatively undervarious attacking scenarios. The results lead to an effective design choice ofsecuring wireless communications for Connected Vehicles. INTRODUCTION: The past couple of years have seen the automotiveindustry teeming with news of R&D towards the development of self-drivingcars.
In fact, Tesla has already conceptualized, tested and commerciallyproduced its autonomous cars – Model S and Model X. With the industry posed ata cusp of revolution, this technology faces ethical and security barriers. Therehas been a case of externals remotely taking control of a Jeep and cutting offits engine while in motion. Should this happen with autonomous cars, theintegrity of the system and safety of the passengers is hugely compromised.
Thus ensuring security is primary concern, for which this paper explores how AIcan help protect communications between connected, self-driven vehicles. AIpredictive algorithms based on Bayes theory like Kalman and Particle Filter areproposed along with generic filters to detect spoofed messages with resilienceto DoS attacks. Utilizing the features unique to surface transportation, thesecurity scheme adopts context adaptive signature verification strategy,significantly reducing the computational overhead in authenticating safetymessages.
Selective validation of beacon messages protects VANET against DoSattacks without losing the effectiveness of faulty message detection. Theresults ensure secure communications for vehicles to vehicles (V2V) andvehicles to infrastructures (V2I). WORK DONE: Architecture of Connected Vehicles Moving cars are nodes havingOn-Board Units (OBUs) which are connected to a network of strategically placedRoad-Side Units in routers. OBUs are special purpose computers made up of twosections – Central Control Unit which acts as a heart of the whole system.
AndHuman-Machine Interface Module which interacts with the person sitting insidethe car. OBUs augment control of the Updated On-Board Diagnostics (OBD-II).OBD-IIs provide electronic means to control engine functions, monitor chassisand accessories including emission, and diagnose car problems.
A Road-Side Unit (RSU)functions like a stationary OBU powered by more computing resources and oftenwith a wired connection to the Internet backbone. RSUs are usually installed atevery 100-200 meters along a road. A spectrum of 75 MHz in the 5.9 GHz band isassigned to the Dedicated Short-Range Communication (DSRC) in VANET. Security challenges in Connected Vehicles The four major security requirements in CVs areconfidentiality, message integrity, end-point authentication, and availability.VANET applications require to use digital signatures, certificates andtimestamps to guarantee message integrity, authenticity, and prevent severalattacks. However, these benefits also come with significant performance costs.
First, digital signatures and certificates increase network messages’ size;thus, they create communication overhead. Furthermore, they add tocomputational overhead. For example, when a beacon message sent and received, asignature is generated per beacon sent and two signature verifications (sendersignature and the certificate authority signature of the certificate) for eachbeacon received. Assuming of 100 vehicles and a beaconing rate of 10 Hz, thisrequires verifying around 2000 signatures per second. When the communication istaken care by beaconing, a vehicle does not know its exact current and futureposition and the speed of other vehicles.
The only information a vehicle has isthe series of beacons transmission rate.Authentication with particle filter People with maliciousintentions can inject bogus messages of traffic jams, or clear roads (when thereactually is a roadblock) to disrupt the traffic. Messages must be authenticatedto differentiate reliable information from fake information. The contextadaptive beacon verification (CABV) security scheme is used along with particlefilter as illustrated in the figure. CABV requires the verification ofsignature from initial beacon of new vehicles and from then onwards checkingthe every nth beacon. To prevent the intermediate spoofed beacons, a linear andnon-linear estimators are used for future position prediction.
If the estimatedpositions and those recorded in the beacon process varies greatly, then asignature is triggered. Simulations were performed using tools such as Sumo,Veins, OMNET++ along with OpenStreetMap before applying the particle filter todetermine vehicle positions, as shown in the figure besides. RESULTS & CONCLUSION: This is the comparison between using Kalman filter andusing particle filter:The above observations concludethat particle filter methodology reduces computational overhead and can providemore accurate vehicle positioning for long distances without using GPS data.However, even a rate of 11% missed spoofed beacons is high for suchlife-threatening applications, thus further work is planned to quantify thesecurity metrics and omit generations, transmission and verification ofsignatures and certificates without significant infringement of security.Parameters, such as spoofed message generation and detection will be definedquantitatively to reduce the undetected spoofed beacon to an acceptable level.
Linkto IEEE Paper: https://drive.google.com/open?id=0B_jVCX29VzNTbVF1VERXak9ZQUE