Automated Drive West: What VSI Discovered on the 2,000+ Mile Drive

2019/10/23

Summary

About This Report

  This report was written by VSI Labs (VSI) for MarkLines' portal users. VSI is a technology research company that provides industry with deep insight and analysis on the enabling technologies used for active safety and automated driving.

 

Contents

  • Introduction
  • Preparation for Automated Drive West (ADW)
    - Dynamic Map Data Loading
        - Downloading and Processing Map Data
    - Map Distortion and the UTM Coordinate System
    - Installation of Human Machine Interface
  • Key Findings
  • Conclusion
  • About VSI Labs

About VSI Labs


VSI Labs reports:
Analysis of Tesla's Over-the-Air Software Updates (Aug. 2019)
The Decomposition of an Autonomous Vehicle Software Stack (Jun. 2019)
Automated Vehicle Ecosystem Analysis (Apr. 2019)
CES 2019: State of the Autonomous Vehicle Industry (Jan. 2019)

 



Introduction

  In August 2019, VSI Labs embarked on a 2,000+ mile cross-country road trip with one of our research vehicles. VSI's vehicle drove from Minneapolis, Minnesota to Santa Clara, California by applying Automated Vehicle (AV) applications on highways.

  The primary purpose of the Automated Drive West (ADW) was to better understand the limitations of over-the-road automation technologies. More specifically, this experiment relied on precision lane models (HD Maps) coupled with absolute localization using real-time kinematic (RTK) corrections.

VSI's Automated Drive West project and research vehicle
VSI's Automated Drive West project and research vehicle


  VSI's research vehicle is a 2018 Ford Fusion equipped with Dataspeed's by-wire control system. The two primary components used in this project were HERE high definition (HD) Live Map and an OxTS inertial navigation system (INS). The OxTS INS is compatible with RTK corrections enabling precision localization. The vehicle was also equipped with a Delphi ESR radar, which was used for adaptive cruise control (ACC). For domain control VSI uses a custom-built Linux-based computer that runs the lane-keeping and ACC algorithms developed by VSI engineers. The vehicle also required LTE connectivity for dynamic map loading and correction services. In essence this was a Level 2+ vehicle because it used an HD map of the road network.

  The ADW was the first experiment in which VSI attempted a cross-country highway drive in AV mode. This article walks through our observations and findings throughout the drive.

 



Preparation for ADW

  VSI used two main AV applications for this drive: HD map-based lane keeping and lane changing, and HD map-based ACC with radar. While VSI has done extensive research on lane-keeping and ACC using HD maps, this research has always been conducted in a geofenced area. In previous tests, VSI engineers downloaded and processed the map data prior to testing. Because the ADW covered more than 2,000 miles of highway roads, it was necessary to develop a method of downloading and processing map data during the drive. In addition, VSI engineers had to adjust our AV system for map distortion. The following sections explains the details of the preparation process.

 

Dynamic Map Data Loading

  Our dynamic map loading algorithm ran and repeated the following steps in order to download and process map data in real-time during the drive.

  1. Download the surrounding map data
  2. Process the surrounding map data
  3. Seamlessly transition from the old map data to the new map data

 

Downloading and Processing Map Data

  VSI wrote scripts to download the correct map data given the vehicle's location at any given time. The script first checks to see if the file already exists, and only downloads the data if it has not already done so. After the new map data is downloaded, the map data must be processed to find a new target path for both lane keeping and ACC. Once a new target path is found using the new map data, the steering command slowly transitions from the current target path to the new target path. Although not strictly necessary, the gradual transition to the new target path allows the driver enough time to react in case any errors occurred in the loading of the new map.

Dynamic Map Loading Diagram
Dynamic Map Loading Diagram: blue tiles are tiles loaded in memory, green tiles are tiles being added into memory, red tiles are tiles being deleted from memory.

 

Map Distortion and the UTM Coordinate System

UTM Coordinate System
UTM Coordinate System
(Image Source: XMS Wiki)

  One of the major difficulties with using maps over large distances is dealing with map distortion. While distortion is negligible over a small geofenced area of a map, the distortion can seriously disrupt map-based lane keeping when relying on centimeter-level accuracy of the map. VSI opted to use the Universal Transverse Mercator (UTM) coordinate system for the Automated Drive West to solve the distortion issues.

  A significant challenge to using the UTM coordinate system is maintaining smooth control when transitioning between zones without disengaging. The UTM coordinates wrap around from the minimum easting value to the maximum easting value when crossing the western zone boundary. Due to this characteristic of the UTM coordinate system, a method is needed to immediately and simultaneously switch the map data and the vehicle's position into the new UTM zone when crossing the boundary.

  To accomplish this, some processing must occur prior to reaching the zone boundary. When the vehicle is within two HERE tile widths of a UTM boundary, any new map data that is being processed will be converted into UTM coordinates for both the current UTM zone and the adjacent UTM zone. As soon as the vehicle crosses the UTM zone boundary, the map data will be overwritten by that of the new UTM zone, the target path will be found in the new map data, the vehicle's location will be converted into UTM coordinates of both the new UTM zone and the previous UTM zone, and a separate orientation offset will be calculated for each location. The steering command will be calculated as a linear combination of the trajectory error between the previous zone's target path with the previous zone's location and orientation, and the trajectory error between the new zone's target path with the new zone's location and orientation. It is not strictly necessary to slowly adjust the steering command from the old zone to the new zone, but it gives the driver ample time to disengage and take control of the vehicle if something were to malfunction.

 

Installation of Human Machine Interface

  To prepare for this drive, VSI developed a new human-machine interface (HMI) device that communicates with the vehicle's domain controller. The device was set up to not only display different outputs regarding the vehicles control commands such as the steering angle and target speed, but also receive certain input from the driver.

VSI Labs HMI device
VSI Labs HMI device installed in the research vehicle


  The HMI device gives the driver the ability to instantly adjust parameters such as the maximum speed for ACC or and center offset for the vehicle's orientation during the ADW, providing a comfortable and smooth ride without unnecessary disengagements. Since the HMI device is configurable to our needs at the time, it rapidly sped up the debugging and testing process for ADAS functionalities such as lane-keeping and ACC while preparing for this project.

 



Key Findings

  Overall, our HD-map based lane-keeping and ACC worked well throughout the drive. We observed that the HD-map based system performed adequately even when lane markings on the highways were not good for much of the drive, making a vision-based system inadequate for sustained lane keeping.

  VSI practices strict discipline when testing automated driving functions and only operates when conditions are safe. We did not enable automated features under dangerous road conditions or conditions where the safety driver would not have adequate time to react. Such conditions include winding roads, low visibility, high roadways without guardrails, areas with pedestrian traffic, and late-night driving. Additionally, the system was disengaged in areas such as highway exits, construction zones, newly constructed roadways, areas with limited connectivity and RTK base stations.

  While many of the observations along the drive were expected due to limitations of the system, key findings from this project are as follows:

  The vehicle's automated features were engaged for 1,767 miles of the 2,000 miles. Most of this was on the Interstate Highway System, a network of controlled access multilane highways in the United States.

  The quality of the HD Maps for the U.S. Interstate Highway System is quite good yielding very positive results with the VSI vehicle. On roads other than controlled access routes, the map data was not good enough to support highway automated lane keeping so we resorted to map-based ACC with a surprising side benefit - VSI discovered that applying maps to ACC improves the performance of radar-based ACC applications because you can filter out extraneous radar points (false positives) that yield inconsistent performance.

  Maintaining connectivity over the western United States was a challenge. Wireless outages limited correction services at times. However, the OxTS device was able to maintain organic accuracy of 30cm, which was enough to bridge large gaps or a falloff in connectivity for periods of time.

  When overtaking semi-trucks, human drivers naturally keep to the far side of the lane to give the truck more room and limit aerodynamic pull from the truck. Our AV was not programmed to do this, and our control stack did not take this disturbance into account. The ability of the control system to cope with large disruptions in airflow was very limited.

  The best performance from the autonomous system took place on Interstate 90 in South Dakota and Interstate 80 in Utah and Nevada. This is where the map data was accurate for the longest periods of time, oftentimes allowing the system to stay enabled for over 50 miles at a time.

  The radar in the front of the vehicle accumulated a lot of dust, bugs and debris along the drive. The output from the radar was unaffected, but other external sensors such as camera and LiDAR would see greater effects from this. As such, sensor cleaning will be necessary for autonomous vehicles driving long distances in the future.

  Keeping AV computers cool when running in warm conditions was also a challenge on the drive west. To cope with this, VSI installed rack mounted server fans to cycle cool cabin air into the trunk where most of the AV gear is located. If we had not done this, the temperature would exceed 110°F and damage the computers. Cooling mechanisms are necessary to maintain proper operating temperatures for autonomous vehicles.

  Lastly, throughout this drive, we reaffirmed the importance of safety driving when testing autonomous vehicle technologies on public roads. The safety driver must always be alert and understand how to use the system, where the system will perform well, and most importantly when the system might fail. Safety driving is more mentally taxing than regular driving, as the driver must constantly anticipate what might happen next. During the trip, it was important to take breaks from driving every few hours, even just to stop at a fuel station.

  For our regular testing, we normally have two people in the car: one safety driver and one engineer. However, for this trip, there were three people in the car to ensure safe testing including the safety driver, the engineer, and the third monitor who sat in the passenger seat and managed the HMI controller to set and adjust desired speeds, adjust lane positions, and monitor the trunk's temperature.

  A lot of communication was required between the safety driver and the engineer. The engineer was watching the lane model to make sure it aligned with what the driver was seeing on the roadway, monitoring the level of position accuracy from the INS device, watching the radar readings to see whether the radar was identifying the closest in-path vehicle for ACC, and finally making sure that the maps were downloading quickly enough. The driver communicated when things became uncomfortable and needed to judge when to ask for the status of certain systems so that the engineer could assist in determining whether to engage, stay engaged, or disengage.

 



Conclusion

  After more than 2,000 miles of exploring the capabilities of lane-keeping and ACC with precision lane models and precision localization, VSI gained valuable insight and data to better understand the limitations of these systems. Testing these map-based solutions in isolation showed the strengths and weaknesses of the system, which is the vital first step in developing redundant systems. The Automated Drive West provided much-needed variety in the road conditions and situations experienced by the VSI research vehicle, giving insight that would not be possible to obtain when driving in a geofenced area with known conditions.

 



About VSI Labs

  Established in 2014 by Phil Magney, VSI Labs is one of the industry's top advisors on AV technologies, supporting major automotive companies and suppliers worldwide. VSI's research and lab activities have fostered a comprehensive breakdown of the AV ecosystem through hands-on development of its own automated vehicle platform. VSI also conducts functional validation of critical enablers including sensors, domain controllers, and AV software development kits. Learn more about VSI Labs at https://vsi-labs.com/.

How to Engage with VSI

  VSI Labs offers subscription research packages to meet your needs:

  • VSI Insights - High level technical analysis of CAV technologies and the future of automated driving.
  • VSI CAV Technology Databases - Deep insights on the products and technologies that make up automated driving.
  • VSI Pro - Decomposition of an AV's functional domains and time saving instructions on how to build an AV.

  Learn more about our portal services or contact us to get started!

 

Conditions of this Report

  In no event shall VSI Labs (a.k.a. Vision Systems Intelligence, LLC.) be liable to you or any other party for any damages, losses, expenses or costs whatsoever (including without limitation, any direct, indirect, special, incidental or consequential damages, loss of profits or loss opportunity) arising in connection with your use of this material. The information in this report is assembled on a best-efforts basis and VSI Labs shall not be responsible for errors or omissions and any claims arising from this.


------------------
Keywords
VSI, autonomous driving, automated driving, autonomous vehicle, Cruise Control, Lane Keeping, Car Navigation System, HD map, Camera, Radar, Sensor, ECU, USA, Ford Fusion, Delphi

<Automobile Industry Portal MarkLines>