Robust CACC through System Adaptation Dr James Martin
Robust CACC through System Adaptation Dr James Martin (jmarty@Clemson. edu) Associate Professor Networking Lab (netlab), School of Computing Clemson University Team: Dr James Westall, Dr Long Chen, Dr Hongxin Hu Students: Anjan Rayamajhi, Manveen Kaur
Overview 1. Network Performance measurement studies of DSRC 2. Implications of network reliability on CACC 3. Heuristic solutions to improve traffic flow adhering to stability and safety 4. Summarize related prototyping efforts - SC-CVT and TGIF 5. Introduce ideas moving forward in areas related to smart infrastructure support for emerging application systems 2
Work related to CACC • Student dissertation “Exploring Smart Infrastructure Concepts to Improve the Reliability and Functionality of Safety Oriented Connected Vehicle Applications” • DSRC performance study • CACC modeling • Techniques for CACC system optimization • Students have contributed to performance analysis studies of DSRC • Cooperative Adaptive Cruise Control (CACC) is advantageous because it: • improves highway throughput • reduces energy consumption • not intended to be a safety application however if we assume autonomous vehicles it is tightly coupled to the x-y control of a vehicle. 3
Problem Statement • CACC systems and its engineering is not well understood for large scale deployments [Naus 2010] [Ploeg 2011] [Ploeg 2014] • Performance and Reliability of applications like CACC in actual DSRC based CV systems is not well understood [Lei 2011] [Ploeg 2015] • Performance when subject to malicious behaviors is not well understood [Biron 2017] • Research challenges: • Can CACC be made more reliable under realistic network conditions ? • Can we generalize an application like CACC and explore the role of smart infrastructure? 4
CACC Contribution • We show that platoons are affected by outages in wireless networks and fall back mechanisms that gracefully move the system to a cautious ACC operating mode are required. • Our methodology introduces a set of CACC system metrics that assess system stability and time to crash events. • We develop and evaluate practical methods by which a platoon can adapt to maximize the vehicular flow rate while ensuring stability. • We have developed both a centralized and a distributed approach 5
Network Performance Metrics • 6
Observations made in DSRC tests 7
Observations made in DSRC tests Calculated Nominal transmission time Observed End to End Latency 8
Cohda OBUs Dual-PHY mode testing Send side: Test Specifics : Rx side Two instances of a WAVE application Channel 172 on Radio A, operating as CCH using Antenna 1 One instance of a WAVE Send rate 10 messages/sec, 184 byte messages Channel 182 on Radio B, operating as SCH using Antenna 2 application Modulation/coding: 1/2 QPSK, queue priority: best Both OBUs running a modified version of Cohda’s WSM-Ech Msgs are tagged with a flow ID – effort to include vehicular data as payload. the server side uses this to Inter-packet Arrival times observed at OBU 2. separate the two streams. No packet loss occurs Note: Cohda does not allow concurrent IP operation
Additional one-way delay results. . .
DSRC Response in Congestion (NS 3 analysis) • How does DSRC behave in a congested scenario ? Number of packets Packet Error Rate • Two nodes with dedicated DSRC flows (Rate ½ QPSK Modulation) • Variable number of nodes in the vicinity • All nodes transmit at 40 pps. • Simulation area of 200 x 200 meters Burst Loss property in Congested DSRC Network Number of nodes 11
DSRC Response in Malicious attack • How does DSRC behave in an attack ? Number of packets • Total nodes = 51 • Two nodes (node 0, node 1) with dedicated DSRC flows (Rate ½ QPSK Modulation) • An attack node (node 50) placed midway between the two nodes • Attack node allowed to transmit at nominal rate • Simulation area of 200 x 200 meters Burst Loss property in attack scenario Modulation and Coding 12
Overview 1. Network Performance measurement studies of DSRC 2. Implications of network reliability on CACC 3. Heuristic solutions to improve traffic flow adhering to stability and safety 4. Summarize related prototyping efforts - SC-CVT and TGIF 5. Introduce ideas moving forward in areas related to smart infrastructure support for emerging application systems 13
CACC Platoons • A platoon of N vehicles in single lane traffic composed of automated trucks • Each vehicle capable of V 2 V or V 2 I using DSRC • Platoon uses information only from vehicle ahead • Each vehicle equipped with RADAR or Li. DAR • Each vehicle undergoes CACC calculations to compute target acceleration • Target acceleration is used by vehicle torque module to actually move the vehicle CACC capable vehicle 14
CACC platoons • Platoon Controller equation CACC Platoon • Vehicle mobility (piecewise linear) 15
How to define Stability in CACC ? • 16
Leader vehicle acceleration Step profile Sinusoidal profile Linear profile Us 06 profile Clemson-Anderson (real) profile 17
Simulating burst packet loss • Two state Markov model No Loss state Table A 18
Simulating a Malicious Node • A vehicle is added in the next lane at the middle of the platoon • Attacks 1. Do. S/jamming: broadcast large frames using a modulation/coding that consumes the most time. 2. Learn the MAC address of all vehicles in the group. Send one unicast packet to each vehicle. The attacker can adapt the rate and the set of targets to obtain the desired impacts. 3. Broadcast WAVE messages at high rates that have 1)small protocol message errors; 2) invalid numeric values; …. • We’ve focused primarily on #1 19
CACC Assessment Metrics • 20
Simulation parameters Configurable system parameters 21
CACC Stability Analysis CACC – no impairment CACC – Do. S Attack ACC- no impairment 22
Time headway vs network reliability – based on sim results N = 10, reliability calculated every 1 sec 23
Time headway vs network reliability vs N 24
Overview 1. Network Performance measurement studies of DSRC 2. Implications of network reliability on CACC 3. Heuristic solutions to improve traffic flow adhering to stability and safety 4. Summarize related prototyping efforts - SC-CVT and TGIF 5. Introduce ideas moving forward in areas related to smart infrastructure support for emerging application systems 25
Dynamic Headway Optimization Control channel Prediction Where, Optimization framework 26
Global Headway Controller • Global controller node located in • An infrastructure node with reliable network connectivity, or • A platoon vehicle with reliable coverage from all the platooning vehicles • Each vehicle submits periodic reliability and mean absolute jerk metrics to the Global controller node using a control channel. • Homogeneous assignment of time headway • The control channel is independent of platoon broadcast channel 27
Local Headway Controller • Local controller located in every platooning vehicle • Each vehicle notes the reliability and mean absolute jerk metrics and decision is made locally • Heterogenous assignment of time headway 28
Results: Simulation settings • Simulation parameters • Simulation loss process setup • 0 – T/3 = No packet loss • T/3 - 2 T/3 = burst loss • 2 T/3 – T = No packet loss 29
Results: Dynamic headway 30
Results: Flow Rate 31
Overview 1. Network Performance measurement studies of DSRC 2. Implications of network reliability on CACC 3. Heuristic solutions to improve traffic flow adhering to stability and safety 4. Summarize related prototyping efforts - SC-CVT and TGIF 5. Introduce ideas moving forward in areas related to smart infrastructure support for emerging application systems 32
SC-CVT • 1. 3 miles along Perimeter Road of DSRC and LTE coverage • Standards based WAVE application testbed • Vehicular nodes: • General purpose CPU, dongles for additional wireless, OBU • Edge nodes: • General purpose CPU, dongles for wireless, RSU, fiber backhaul to campus • Planned: • Video camera at each Edge node • Experimental testbed • WAVE applications not tied to a networking stack or specific spectrum
TGIF • Idea: community infrastructure to promote domain research ‘out in the wild’ • Two main components • Campus deployment: make it easy for researchers/staff/students to deploy Io. T devices (any wireless device). • Any Linux device can utilize our middleware • Other devices interface to TGIF through gateway nodes • Message dissemination based on publish/subscribe - system consists of edge and system nodes running MQTT brokers. • Backend database and portals for data analysis: currently exploring Elastic database
Overview 1. Network Performance measurement studies of DSRC 2. Implications of network reliability on CACC 3. Heuristic solutions to improve traffic flow adhering to stability and safety 4. Summarize related prototyping efforts - SC-CVT and TGIF 5. Introduce ideas moving forward in areas related to smart infrastructure support for emerging application systems 35
Citizen’s Wireless Internet The FCC wants under utilized spectrum to be designated as ‘shared’. -DSRC spectrum has been tagged by the FCC as ‘shared spectrum’ -Two proposals 1. Requires Wi. Fi to sense the presence of an incumbent (and to immediately stop using the spectrum) 2. Eliminates the lower DSRC channels Note: The CBRS spectrum (3. 5 GHz) is likely the first block of shared spectrum that will become available to established operators or innovative companies.
Citizen’s Wireless Internet • The overall goal is to develop innovative techniques that better support the wireless needs for emerging application systems such as connected vehicles. • One research focus builds on dynamic spectrum sharing in a multi-RAT hetnet with centralized coordination of participating wireless nodes. • The level of cooperation of the underlying component wireless networks is an exploratory topic in the research. No cooperation assumes each wireless network in the geographic area of interest is effectively a black box. 'Some' cooperation might map to each wireless network provides minimal information. • Further cooperation (and one of our main ideas) between independent networks can be achieved through dynamically managed shared spectrum. Our proposed system extends the FCC's current CBRS/SAS architecture that would interact with a variant of our ongoing intra-RAT hetnet system. • Application System: a collection of applications that operate in a distributed manner that share some level of commonality Classification of Application Systems: 1. Mobile Wireless Network Application 2. Custom Application System (UAV Swarm, Io. T Smart City) 3. CWI Command Control Application 4. Critical Infrastructure V 2 V 5. Critical Infrastructure Custom Application System (Public Safety, Smart Grid)
CWI System Model 38
Low Latency Networking • The Internet community has always had a research thread on the topic • Resurfaced the last 10 years initially due to ‘buffer bloat’ and more recently due to Io. T and ideas similar to the application systems topic. • There’s a move to extend the Internet’s single best effort service model to a two level service model: low latency and best effort. • Techniques involving Active Queue Management and Explicit Feedback Notification are involved • The Io. T community has specific scenarios such as smart grid, manufacturing processes, machine-to-machine interactions including connected vehicles. • The most recent Wi. Fi standards update specifically addresses low latency – wifi chipsets will be able to perform localization functions in addition to optimized protocol enhancements • The most recent 5 G standards have defined support for Ultra-reliable low latency communications (URLLC) 39
Conclusion • We have shown that CACC can be affected by wireless congestion and malicious attacks making fallback strategies and ‘smart infrastructure’ critical • CACC (and future cooperative AV’s) are examples of ‘Application Systems’ that are a part of our Nation’s critical infrastructure. • Many emerging Application Systems require wireless infrastructure • The government is clearly conflicted • There is a recognized need for isolated spectrum and systems (consider public safety’s first. Net) • But there’s a strong movement to keep the government out of it • This motivated our idea for the CWI – an incremental enhancement to the FCC’s dynamic spectrum mgmt. direction 40
References: [Naus 2010] G. Naus, R. Vugts, J. Ploeg, R. v. d. Molengraft, and M. Steinbuch. Cooperative adaptive cruise control, design and experiments. In Proceedings of the 2010 American Control Conference, pages 6145{6150, June 2010. doi: 10. 1109/ACC. 2010. 5531596. [Ploeg 2014] Jeroen Ploeg, Nathan Van De Wouw, and Henk Nijmeijer. Lp string stability of cascaded systems: Application to vehicle platooning. IEEE Transactions on Control Systems Technology, 22(2): 786{793, 2014. [Biron 2017] Zoleikha Abdollahi Biron, Satadru Dey, and Pierluigi Pisu. On resilient connected vehicles under denial of service. 05 2017. [Ploeg 2015] J. Ploeg, E. Semsar-Kazerooni, G. Lijster, N. van de Wouw, and H. Nijmeijer. “Graceful degradation of cooperative adaptive cruise control”, IEEE Transactions on Intelligent Transportation Systems, 16(1): 488{497, Feb 2015. ISSN 1524 -9050. doi: 10. 1109/TITS. 2014. 2349498 [Lei 2011] C. Lei, E. M. van Eenennaam, W. K. Wolterink, G. Karagiannis, G. Heijenk, and J. Ploeg. Impact of packet loss on cacc string stability performance. In 2011 11 th International Conference on ITS Telecommunications, pages 381{386, Aug 2011. doi: 10. 1109/ITST. 2011. 6060086. [Ploeg 2011] J. Ploeg, B. T. M. Scheepers, E. van Nunen, N. van de Wouw, and H. Nijmeijer. Design and experimental evaluation of cooperative adaptive cruise control. In 2011 14 th International IEEE Conference on Intelligent Transportation Systems (ITSC), pages 260{265, Oct 2011. [Rayamajhi 2018] Anjan Rayamajhi, Zoleikha Abdollahi Biron, Roberto Merco, Pierluigi Pisu, James M Westall, and Jim Martin. The impact of dedicated short range communication on cooperative adaptive cruise control. In 2018 IEEE International Conference on Communications (ICC), 2018. [Rayamajhi 2019] A. Rayamajhi, “Exploring Smart Infrastructure Concepts to Improve the Reliability and Functionality of Safety Oriented Connected Vehicle Applications”, Ph. D. dissertation, 12/2019, online: https: //people. cs. clemson. edu/~jmarty/papers/Students/Dissertation_Rayamajhi_Anjan. pdf 41
Thank you 42
Network Diagram (congestion) Propagation and fading model used from ** J. Benin, M. Nowatkowski and H. Owen, "Vehicular Network simulation propagation loss model parameter standardization in ns-3 and beyond, " 2012 Proceedings of IEEE Southeastcon, Orlando, FL, 2012, pp. 1 -5. Receiver 50 m 200 m Transmitter 200 m 43
Network Diagram (attack) Receiver Transmitter 50 m 200 m Attack Node 200 m 44
SC-CVT
Cohda OBU 1 Ch. 172 (Flow 1) Ch. 182 (Flow 2) Cohda OBU 2 46
Traditional ACC Stability Analysis Crastime, sec Crash Analysis Time headway, sec 47
CACC with Malicious attack Stability Analysis Crastime, sec Crash Analysis Time headway, sec 48
CACC with no packet loss Stability Analysis Time headway, sec 49
CACC with complete outage Stability Analysis Crastime, sec Crash Analysis Time headway, sec 50
CACC with Congested network Stability Analysis Crastime, sec Crash Analysis Time headway, sec 51
CACC with On-Off burst loss Stability Analysis Crastime, sec Crash Analysis Time headway, sec 52
CACC with long burst loss Stability Analysis Crastime, sec Crash Analysis Time headway, sec 53
False True Vehicles GHC 54
False True Local Headway Controller 55
Local Estimation (Kalman Filter based) control inputs • Sensor measurements predictio n • Prediction Phase correctio n • Correction Phase 56
Parameters for Acceleration Estimation Jerk Distance traveled Velocity acceleration State variables Measurements 57
Simulation of CACC 58
- Slides: 58