Automated CV simulation tool-chain based on VENTOS for Simulation level validation for multi-node automotive platooning EE architecture under false data injection attack

automated tool-chain has been built on top of an existing connected vehicle simulation tool, VENTOS, for CV traffic. The outline of the proposed tool-chain is given below. The proposed tool-chain can be used to generate sequences of false data (attack vectors) for any type of platoon structure. It also provides a visual simulation of that attack vector on the underlying platoon structure to simulate such an attack on a platoon where i) each vehicle's dynamics can be customized, ii) the platoon topology can be defined. This enables us to investigate the different vulnerable situations designed for different such system dynamics and CV topologies. Customizable vehicle dynamics can be further leveraged for the development of control-theoretic attack detectors.

Attack detection algorithm design for single node automotive EE architecture under false data injection

The ideal target to inject a false data injection attack (FDIA) in an automotive EE architecture is the controller area network or CAN bus because it is used by most of the safety-critical control loops in a vehicle. Lack of any security schemes makes CAN protocol vulnerable to such attacks as well. A message ID that an FDI attacker intends to falsify is called the victim message and the electronic control unit (ECU) that is responsible for transmitting the victim message is called victim ECU. The first step of FDIA is achieved by sending the victim ECU to the CAN bus to bus-off. The following figure demonstrates the phases of bus-off (step 1, 2, 3). Following the disconnection of the actual ECU the attacker is able to send falsified messages that replaces the victim messages.

Computer-Aided Design (CAD) Framework for Simulation level validation for single node automotive EE architecture under false data injection attack

Computer-aided design (CAD) framework has been developed for estimating the vulnerability of automotive CPSs. Outline of the proposed framework is given in the below figure. We have considered a model-based representation of safety-critical automotive controllers and monitoring systems working in a closed loop with vehicle dynamics and verified their safety and robustness with respect to false data injection attacks. The proposed framework tries to find out which sensor and/or actuation signal is vulnerable by generating stealthy and successful attacks using a formal method-based counterexample guided abstraction refinement (CEGAR) process. Software tool for basic attack vector generation of simple linearized automotive control loops has been made ready. We intend to make a structured methodology to visualize the effect of the formally synthesized attack vectors that are designed to violate specific system properties and how to counter them.

Attack detection algorithm design for automotive platooning EE architecture

A preliminary version of a detection mechanism to detect false data injection (FDI) attacks on a platoon has been developed. Each vehicle's state x consists of position (s), velocity, (v) and acceleration (a) They communicate their states to their neighbor vehicles. The tool-chain, described in the previous step, considers an attack model where an attacker falsifies a vehicle's state while the state is being communicated.

Virtual software-in-loop (SIL) testing of adaptive cruise control (ACC) system designed for light-duty vehicles in Carmaker 8.0

In car simulation software Carmaker 8.0, we have built a cruise control model for a single car. We design the controller in Simulink which is connected to the plant in Carmaker. The actuation of control signal on the vehicle's longitudinal dynamics is visualized on Carmaker (Fig. 8) where on input of any constant speed, the controller will make the car move with that specific velocity. This facilitates the design and validates our own controller.

Real-time hardware-in-loop (HIL) emulation of adaptive cruise control (ACC) system (for single node) under bus-off attack and detection in controller area network (CAN) protocol

We first successfully implemented CAN bus-off attack on an Infineon ECU in closed loop with an ETAS Labcar RTPC emulating an ACC system. The attacker is implemented in an Arduino Uno interfaced with the CAN bus using Sparkfun CAN shield. The following figure demonstrates how the closed-loop behaves under the bus-off attack followed by the FDIA. On application of our aperiodic control execution-based detection algorithm, the bus-off attack is eventually detected and the closed loop stabilizes under the aperiodic control execution.

Development of a platoon test-bed that consists of race cars capable of performing autonomous maneuvers

For the purpose of conducting rigorous analysis upon implementation of attack, detection and mitigation algorithms, a test bed has been developed, consisting of four one-tenth scale race cars. Utilising the onboard processors, sensors and actuators, makes it capable of running modern algorithms in the discipline of AI, Deep Learning, Computer Vision and a lot more.

Simultaneous Localisation and Mapping (SLAM)

Adaptive Monte Carlo Localisation (AMCL)

End-to-End Learning

Cooperative Adaptive Cruise Control (CACC)

A vehicle-to-vehicle communication network is created using a python socket library which allows cars to transmit and receive data using the processor's IP address. This is achieved via client and server scripts running as threads running in each car according to the decided network topology. A car sends its state (acceleration, velocity, position, orientation) to the cars connected to it. Based on the state, the control actions are calculated for safe maneuvers in a connected vehicle platoon.

Work In Progress

Work In Progress

Future Plan

1. In case of vehicle platoons, a stealthy attacker is likely to have registered itself as a legitimate member of the platoon. Therefore, certification-based authentication can be fooled. Attacker vehicles can send false data to the platoon members to initiate collision or reduce traffic throughput without creating any suspicions. Threshold-based anomaly detectors are widely used to detect such false data. In this regard, we plan the following two works:


1. We will explore the statistical change detection methods (like, chi-square, CUSUM, etc.) to determine the suitable threshold for the attack detector while considering the distributed nature of attacks in the platoon.


2. Further, we also plan to make the detection more robust against intelligent attackers by framing a two-player game between the attacker and detector. Both the attacker and detector agents learn the environment and each other's strategies using standard RL policies. The objective of the attacker would be to stealthily falsify the communication data to violate safety/performance goals as early as possible. Whereas, the detector's objective would be catching the intelligent attacker before the latter succeeds.


2. In cooperative adaptive cruise control (CACC), the vehicles in a platoon use NN-based image processing along with other sensor data to gather information about their surroundings. With this information, NN-based controllers intelligently maneuver the platoon vehicles. Data falsification attacks can hamper the safety of such controllers. We plan to evaluate the safety of the NN-based controllers in presence of data falsification attacks leveraging reachability analysis. This can serve as a mitigation strategy against data falsification attacks.


3. Our final goal would be implementing the proposed control-theoretic detection and robust mitigation methods on hardware-in-loop real-time setup as well as a small-scale platoon setup using f1/10 vehicle chassis.