robert fenton’s early osu workcitr.osu.edu/research/older-work.pdf · • intersections, traffic...

Post on 14-Jun-2020

1 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

1970s: Robert Fenton

• Longitudinal car-following control using take-

up reel as a stand-in for nonexistent sensors.

• Lateral control and platooning using leaky

antenna cables.

• RADAR sensor development.

• Feedback to driver through actuated joystick.

• All implemented on analog computers

and early microcontroller systems.

ROBERT FENTON’S EARLY OSU WORK

1996-1997: Image processing based lane

tracking and lateral control (WRC)

1996-1997: RADAR reflective stripe lane

tracking and lateral control (VDA)

1997: NAHSC (San Diego) demonstration

prep (Skidpad)- RADAR stripe based lane

change and passing maneuvers

1999: Electronic Tow Bar- image based

vehicle following and convoying (lateral

and longitudinal control)

CITR: THE NEXT GENERATION

AUTONOMOUS DRIVING AT OHIO STATE

Autonomous Driving Areas• Lane tracking

• Car following

• Intersections, traffic circles

• Passing

• Obstacle avoidance

• Parking

• Dynamic route planning

ACT 2007

Demo 1997TerraMax 2004 ION 2005

DARPA URBAN CHALLENGE (2007)

T-junction (Area A) in DARPA Urban Challenge

▪ Fully autonomous driving in a city environment with other vehicles.

▪ Conditions in T-junction (area A)

- Blue and green lines are the path for human drivers, red line is for autonomous

driving vehicle.

- Human drivers passing the junction without stop or adjusting their speed.

- ACT needs to measure the speed of passing cars and find an appropriate gap.

▪ OSU-ACT Area A Qualifier

[Area A of the DARPA Urban Challenge]

DARPA URBAN CHALLENGE (2007)

Design of situations

▪ Implement the Meta-State (sub-states + FSM)

Hybrid State System (OSU-ACT Controller)

▪ Vehicle controller for OSU-ACT consists of a Low-level Controller (LC) and a High-level Controller (HC).

▪ HC is for the conscious-level decisions (e.g., lane-change)

▪ LC is for the subconscious control of steering and throttle/brake.

▪ 𝛹,𝛷, and 𝛤are events generated by HC, LC, and the sensing and analysis system respectively.

Desired paths

feedback

Steering, velocity

Desired paths

(position, velocity)

Threshold check

Command completion

[ Meta-state connections ]

SCALED-DOWN TESTBED ARCHITECTURE

FOR URBAN SCENARIOS (2014-16)

Characteristics

▪ Scale 1/7

▪ Scaled down sensors or virtual sensors

▪ Virtual GPS module and camera

system.

▪ Wireless 802.11b interface for

communication.

▪ Limitations are added by software.

Hybrid-State System

High level

control

Finite State

Machine

Low level

control

PID-based

continuous

controller

[2] Ozbilgin, G., Kurt, A., & Ozguner, U. (2014, June). Using

scaled down testing to improve full scale intelligent

transportation. In Intelligent Vehicles Symposium

Proceedings, 2014 IEEE (pp. 655-660). IEEE.

Implemented application example:

Lane Change Scenario

▪ Mixed traffic environment.

▪ Beneficial for V2V equipped cars

without perception.

▪ Different wireless information sharing

strategies implemented.

SCALED-DOWN TESTBED ARCHITECTURE

FOR URBAN SCENARIOS (2014-16)

[3] Adamey, E., Ozbilgin, G., & Ozguner, U. (2015). Collaborative

vehicle tracking in mixed-traffic environments: Scaled-down tests

using simville (No. 2015-01-0282). SAE Technical Paper.

[4] Ozbilgin, G., Ozguner, U., Altintas, O., Kremo, H., & Maroli, J.

(2016, June). Evaluating the requirements of communicating

vehicles in collaborative automated driving. In Intelligent Vehicles

Symposium (IV), 2016 IEEE (pp. 1066-1071). IEEE.

DEVELOPMENT OF STANDARD LDW/LDP

TESTS AND EVALUATION METHODS (2014)

Impact (2004 data, ~6M incidents)

[5] Kurt, A., Özbilgin, G., Redmill, K. A., Sherony, R., & Özgüner, Ü. (2015). Test Scenarios, Equipment and Testing

Process for LDW LDP Performance Evaluation (No. 2015-01-1404). SAE Technical Paper.

Driver data analysis, driver behavior analysis

▪ Crash databases

▪ Investigating departure scenarios

Simulator studies for data collection

▪ Test procedure design

▪ Identifying important data fields

Test scenario selection

▪ Prioritizing the test matrix

Test/Evaluation procedure development

▪ Driving simulator for procedure design

▪ Vehicle instrumentation for data collection

Vehicle Evaluation

▪ At Transportation Research Center

▪ Two vehicles, roughly five days of testing with each

• Autonomous and semi-autonomous scenarios are being explored.

• Exploring information requirements for inter-vehicle communication

to provide safe and smooth operation.

• Exploring platoon/convoy configuration and control.

• Experimental testing done at OSU

COLLABORATIVE LANE CHANGE/MERGE

(RENAULT/CAR CONSORTIUM)

Lane Change Trajectory Prediction (2014)

• Use the front view camera to predict nearby vehicle’s intension of lane change,

and the future vehicle trace of marge-in maneuver.

• Real driving data: 210 normal lane change instances from SHARP2, and 140

dangerous instances from 100Car near-crash data

[6] Liu, P., & Kurt, A. (2014, October). Trajectory prediction of a lane changing vehicle based on driver behavior

estimation and classification. In Intelligent Transportation Systems (ITSC), 2014 IEEE 17th International Conference.

COLLABORATIVE LANE CHANGE/MERGE

Convoy Control with a Cut-in Vehicle (2015)

• Control of a CACC convoy for nearby vehicle’s cut-in maneuver.

• The convoy host vehicle predicts a vehicle’s lane change behavior, calculates

control of each convoy member, and assigns control by V2V.

• Optimal control is calculated to minimize speed deviation and headway

fluctuation, benefiting fuel economy, traffic capacity, and spacing safety.

[7] Liu, P., & Özgüner, Ü. (2015, July). Predictive control of a vehicle convoy considering lane change behavior of

the preceding vehicle. In American Control Conference (ACC), 2015 (pp. 4374-4379). IEEE.

COLLABORATIVE LANE CHANGE/MERGE

Distributed MPC for Flexible Vehicle Platooning (2016-17)

• Control method for cooperative vehicle platooning in automated highway systems,

allowing new vehicle’s merging-in for flexible convoying.

• A decentralized MPC method partitions the convoy into clusters, and solves the

control problem of accepting new vehicle in a non-iterative way based on V2V.

• The control method is proven to

guarantee convoy string stability

and collision-free safety.

[8] Liu, P., & Ozguner, U. (2017, May). Non-iterative distributed model predictive control for flexible vehicle

platooning of connected vehicles. In American Control Conference (ACC), 2017 (pp. 4977-4982). IEEE.

[9] Liu, P., Kurt, A., & Ozguner, U. (2018). Distributed Model Predictive Control for Cooperative and Flexible Vehicle

Platooning. IEEE Transactions on Control Systems Technology.

COLLABORATIVE LANE CHANGE/MERGE

OTHER WORK

SMOOTH Smart Mobile

Operation: OSU Transportation Hub

Most people live or work far

from a transportation stop. Network of “On-

demand automated

vehicles”Transportation stops not

close to points of interest.

DURA Automotive Automated Valet (2015-2016)

• Fully autonomous navigation in parking lot

• Automated head-in, tail-in and parallel parking.

• Vehicle DBW conversion, path planning, sensor-based localization, vehicle control.

Partial Automation and V2X

• Using a longitudinally-automated vehicle to demonstrate V2X potentials.

• Intelligent traffic light passing for fuel economy.

• Automated decision on intersection precedence at stop signs for safety.

Ongoing Work

Driving

Characteristic

or

Driving style

Driving

Events

Environ-

mental

Conditions

Goal of

Driving

Events

𝐹𝑆𝑀 & 𝐸𝑀

Traffic

conditions

[10] Jing, J., Kurt, A., Ozatay, E., Michelini, J., Filev, D., & Ozguner, U. (2015). Vehicle Speed Prediction in a Convoy

Using V2V Communication. ITSC, 2015–Octob, 2861–2868. https://doi.org/10.1109/ITSC.2015.460

Driver Intention Prediction for Car-following and Lane Change

▪ Analyze the driving patterns by using real-traffic data (US-DoT NGSIM traffic data).

▪ Determine the type of driver and pattern of maneuvers.

▪ Investigating a hybrid model to detect the intention and predict the future state:

- Estimation of a deterministic driver model + Training of a probabilistic model.

Ongoing Work

Autonomous Parking & Docking of Tractor-Trailer Vehicles

▪ A control-oriented model and a physical based model.

▪ Three path tracking control methods by PI, Sliding Mode, and Neural Network.

▪ A generic jackknife accident prevention system for all path tracking controllers.

▪ A decomposed method by Markov Reward search and QP optimization.

▪ Team collaboration results: 2 papers and 1 patent application, all within 6 months.

top related