We validate our proposed mitigation algorithm on a platooning testbed and compare it with the baseline algorithm i.e., regular platoon control law with integrated collision avoidance.
2.5.1 Hardware Setup
Our experimental setup consists of small robots that represent vehicles in a stream of platoons and a motion capture system for tracking as shown in Fig. 2.4. We implemented the disbanding attack and the traveled distance mitigation algorithm on three 3-vehicle platoons, denoted as per the convention shown in Fig. 2.2. The 3rd (leading) platoon
disbands and the response of other two platoons is captured with the different algorithms in place.
Each robot is affixed with multiple IR markers which are tracked by the Optitrack motion capture system consisting of 24 IR cameras and the Motive software that enables us to capture the robot positions. This position data is then streamed to a command computer where an interface application utilizing the Robot Operating System (ROS) [35] framework makes the gathered position data for each robot available to our controller application. This
Fig. 2.4. Experimental environment with small robots and motion capture system
application processes the position data and sends control commands accordingly to each robot. The controller application implemented on ROS works in the following manner:
• The raw position data is processed using an Extended Kalman Filter to reduce camera sensor noise and estimate the measured position and velocity.
• Pure Pursuit Controller utilizes the extimated positions and circular path coordinates from the experiment environment to calculate the angular velocity command for each vehicle.
• The estimated data of all vehicles is used to calculate the relative distance and ve- locity between consecutive vehicles. This is then fed to a High level Controller which implements the platoon model following the bidirectional control law as explained in Section2.2.1and provides desired acceleration values for the robots.
• The mitigation and baseline algorithm then modify the acceleration values from the High level Controller in case a disbanding attack is detected.
• As the vehicles only act upon instantaneous velocity commands, these acceleration values along with current measured velocities are used to calculate the desired veloc- ities for each vehicle. The desired linear velocities for the vehicles are achieved using a PI controller which acts as our Low Level Controller. This controller calculates the linear velocity commands for each vehicle such that the measured and the desired velocities match.
Each robot consists of a 32-bit ARM-based mbedNXP LPC1768 microcontroller on the Pololu m3pi platform to which the Digi Xbee receivers are interfaced. The corresponding Xbee transmitter is connected to the command computer. These Xbee modules allow us to establish a wireless communication channel using the Zigbee protocol over which the angular and linear velocity commands calculated for each robot using our controller application are then broadcast. The firmware on these robots receive the broadcast messages and calculate the left and right wheel speeds from the received angular and linear velocities as per the differential drive model.
2.5.2 Experimental Results
Fig. 2.5 shows individual velocity profiles for the vehicles under consideration (three platoons with three robots in each). Fig. 2.5a indicates the effect on velocity due to disbanding for the baseline control structure, given in Section 2.2.1, wherein we can see vehicles in the last platoon not only slow down suddenly, but one of them stops, in response to the disbanding of the lead platoon. Fig. 2.5b and 2.5c give the velocity profiles when intact robots use the traveled distance mitigation approach, wherein it can be seen that the speed of vehicles in second and third platoon slow down gradually and then begin to accelerate. This mitigation approach was tested with ts = 0.5s and 1s, respectively. The
point labeled as A in Fig. 2.5a, 2.5b, and 2.5c indicate that the platoons are in a steady state. Point B marks the time at which the attack on the lead platoon is emulated, causing all of its vehicles to disband and suddenly decelerate. Deceleration patterns of the vehicles after point B for the baseline structure clearly indicate a sudden drop in velocities for the
0 5 10 15 20 25 30 35 Time [s] 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Speed [m/s] Vehicle 3-1 Vehicle 3-2 Vehicle 3-3 Vehicle 1-3 Vehicle 1-2 Vehicle 1-1 Vehicle 2-3 Vehicle 2-2 Vehicle 2-1 A B C (a) 0 5 10 15 20 25 30 35 40 Time [s] 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Speed [m/s] Vehicle 3-1 Vehicle 3-2 Vehicle 3-3 Vehicle 1-3 Vehicle 1-2 Vehicle 1-1 Vehicle 2-3 Vehicle 2-2 Vehicle 2-1 A B C (b) 0 5 10 15 20 25 30 35 Time [s] 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Speed [m/s] Vehicle 3-1 Vehicle 3-2 Vehicle 3-3 Vehicle 1-3 Vehicle 1-2 Vehicle 1-1 Vehicle 2-3 Vehicle 2-2 Vehicle 2-1 A B C (c)
Fig. 2.5. Vehicles’ velocities upon disbanding of platoon 3 for baseline control structure and proposed heuristic mitigation algorithm with ts = 0.5s and 1s.
following platoons, causing some vehicles to come to a complete stop as indicated by point C.
While there are no collisions with the baseline control, sudden deceleration/ accelera- tion was observed. However, such abrupt changes in velocities are not observed when our proposed heuristic mitigation is in place (see Fig. 2.5band 2.5c, where point C shows that none of the vehicles need to come to a halt). With the mitigation approach, vehicles com- fortably decelerate and gradually accelerate to recover and maintain desired spacing and velocities, all without collisions. Furthermore, Ev was calculated for the three experiments
and it was equal to 30.02%, 21.82% and 19.73% for Fig. 2.5a,2.5b and 2.5c, respectively. These numbers indicate that with increasing ts, the change in velocity is even smoother
and more gradual, yet collisions do not occur. However, with ts = 1 s, vehicles come to
uploaded short videos of our experiments [36].