Autonomous driving (AD) is an interesting development but it will not be available any time soon. We can see two roads leading to complete autonomy: the 'Google Route' with a final goal of having complete autonomous vehicles on the roads. The other, a more conservative road, leads us through the expansion of advanced support systems for the driver, also referred to as advanced driver assistance systems (ADAS). These systems are gradually taking over more and more driving tasks until the day comes that driving the vehicle is only something that concerns the vehicle. Perhaps the final step will not be that big a deal - simply allowing for complete autonomy to happen.
Article from Objective #24, November 2015
As an organization we have been interested in AD and ADAS for quite some time. We have already started using a number of ADAS systems in practice. However, it still requires a big investment to purchase a number of vehicles, add the control systems and then find an area big enough to allow them a test drive.
Our R&D project
Almost a year ago, we started an informal project with one simple goal: “We want to race self-driving model vehicles in October 2015 at the ITS World Congress” … it's as simple as that. The ITS World Congress 2015 was to be held in Bordeaux in France, a city along the river Garonne. This is how the Garonne project came to be, with a simple goal and a harsh deadline.
The first six month of the project were purposefully (relatively) unstructured, with volunteers that each choose an aspect of the problem they were interested in and spent their free time working on it. Each research orientated activity like Garonne requires a lot of freedom, which forced us to put most of our usual development rules aside. Except when they were safety-related. We have been constructing Wiki's and Forum systems in order to share information, a git repository to share codes and documents and we have also made a lab for this project. Then we went to work; or rather, we went home to go to work.
The result was stunning; technical experts explored each topic, from 3D printers to optical positioning systems, from RF communication to planning systems and from embedded control systems through simulators to new source code control mechanisms. The biggest challenge during this phase was to keep track of everything that was going on, to understand what colleagues were working on and what belonged in the wider Garonne vision. It was a fantastic experience. Practically every day, the team has something new to share. Our first 3D printed cogs, messages that can be sent through wireless connections or motor controls that made the wheels turn for the first time - there was a constant flow of new developments. Small successes that would inspire one another - no one wanted to disappoint the team and be the reason the demo would not be finished in time for Bordeaux so the progress was amazingly quick.
Some examples of what we have developed:
- A complete, modular, 3D printed, configurable vehicle chassis: When we started with the project we had limited experience with 3D printers but the capacity to quickly experiment and try out new designs were extremely useful. Not to mention the cost reduction when you can simply print interchangeable parts when two vehicles have crashed into each other.
- A ‘Fly By Wire’ four-wheel drive, four-wheel drive control platform: Each wheel has its own control system that determines speed, position and force. The communication with the wheel goes through a trusted CAN bus.
- Vehicle dynamic as software: the independent wheel drive allows for the control, propulsion and brake function all to be implemented in the software. Locking differential, independent steering, active correction and ABS were all software modules instead of complex mechanical constructions.
- Infrastructure low-level control: software control over a vehicle requires a platform that can respond reliably and in real-time. In order to meet this requirement, we have developed a platform with a control system with interfaces for each individual vehicle component.
- Positioning the interior: we soon found out that a GPS cannot offer any help in a demonstration area so we developed a location infrastructure that uses infrared 'light beacons'. This provided precision up to 20 mm inside the operational space.
- V2V- and Vehicle-to-Infrastructure-communication: for a number of projects we use ITS-G5 for V2V/V2I, but this is a little too much of a good thing for model vehicles. That is why we developed our own communication system based on the same principles that help run G5 but on a 868MHz ISM band.
- Cameras with 3D view and depth of field based on time-of-flight technology: This promising technique creates point clouds that show the distance to objects within a field of vision. The cameras wer used only have reach of a few meters but there is equipment that reaches even further.
- A real-time 3D simulator based on the USARSim motor: This allowed us to make great progress with the high level software while the low level and physical systems were still in their development stage.
There have been some important lessons that taught us a lot. We realized that 3D printing is very useful during the prototype and research phase of a project. Certainly when mechatronic subsystems are involved. The quality you can get is perfectly acceptable for mechanical components and prototypes; the reliability and the physical consistency of the materials were never a problem.
The 'Fly by Wire' wheels make a lot of space in the vehicle that can be used for electronics and power systems while the software dynamics offered great flexibility regarding the controls of the vehicle. Especially the implementation of the special low level controls was a smart move, because low level controls of the vehicles redirected us from the high level 'thinking processes' concerning certain goals. Communication between the low level and the high level platforms went through an Ethernet connection and offered extra flexibility to control the vehicles with WiFi during the development.
To our surprise, the process was not expensive. Obviously, it required a few thousand Euros for 3D printers, plastic, cameras, parts and engines, but the fact that people worked in their spare time completely voluntarily was budget-friendly and made sure the symbioses of this project was quick to show. The participants were allowed to play with interesting new technologies without there being any pressure; the business and participants had fun, were given a personal experience and obtained research and training as well as a better understanding of technologies that could be applied immediately in our client projects.
Not that there were a lot of failures. In a 'normal' project, a big part of the activity is invested into risk management. In other words, to identify and limit any risks that are inherent to any developing process - from start to finish. But when the motto is: 'Let's see what happens' or 'we can try', development becomes that much easier. One of the things that was clear was that not a lot of things went wrong, despite the fact that there was no desire for any safety net and we deliberately took risks. Of course, we adjusted direction and the learning process occasionally. This often resulted in the fact that we also had to adjust our approach. But there was no part in the entire program that did not make it to the finish line.
Switching to project management
In March 2015, we have established what we had in order to be able to see which parts were usable for the general goal of the Garonne demo and for our original vision. More important was that we established which pieces of the puzzle were still missing due to the ad-hoc strategy we had been using.
The result was not disappointing. What was missing was a sensor infrastructure and high level integration. Still crucial: the vehicle was (not yet) fully autonomous. At that moment, Garonne became a project that would run during office hours in order to switch from research to applied research and make a prototype for the demo. Genuine projects need project managers and that is why we used our usual structures to get started.
You would probably expect that that would be that. Towards the end of September, the demonstration system was ready for use; everybody was happy. But it was not that simple. The switch from 'office hour project' was made right before the summer (vacation time) so we had to go without certain resources and this resulted in a huge planning problem. In addition, the challenges that presented itself were concerned with the delivery of hardware, sensor integration, acceptance of the 3D print techniques that were being used and a few more issues. Only at the beginning of September could we start some autonomous aspects for ourselves. Luckily, our normal control processes were in place because we had since switched to Technolution project modus. That is the difference between research and development. Research can comply to accommodate various tools, techniques, people power and knowledge without losing any of its usability but during development there is a clearly defined and specific result to be had; you will have to design the process in such a way that you can make your goal, while keeping the limitations of reality in mind. This is a crucial difference between a university research group and a system integrator that is functioning ahead of the game.
And? Did we succeed? Absolutely! The demo at the ITS World Congress was impressive. The four vehicles did not race, they drove around and worked together, but despite that, the demo was complete and generated a lot of interest. Of course, there were some accidents in September, because we could integrate autonomical aspects in the vehicles around that point. But even those crashes could be transformed to become improvements. We could therefore experiment with model vehicles, crash some and learn from it. This is harder in reality where each vehicle costs at least 20,000 Euro and weighs 1,5 tonnes. We are more interested in driving 5 km and learning from 20 accidents than driving a million km without learning anything.
How to proceed? At the World Congress, we were approached by a few organizations that were interested in our vehicles for their own experiments with autonomous driving and we are seriously considering to take them up on their offers. Do you wish to known what we did exactly? Download the paper that we wrote for the ITS World Congress.