Jónsson's, of Reykjavík University, spoke at the 2010 Driving Sustainability conference in Reykjavík, Iceland last weekend on the topic "Ultimate Sustainability." In this case, ultimate isn't an understatement. Jónsson helped NASA develop the electric rovers, Spirit and Opportunity, that were sent to the Red Planet in 2003 and performed way beyond anyone's expectations since arriving in the harsh, harsh environment. Think Nissan is being careful with the cold weather package for the Leaf? Try getting a battery ready for temperatures that can drop to -50 or -80 C at night. Then try powering these batteries from the sun in a place that gets less solar energy (Mars is further from the sun than the Earth) and where the sky is often covered by dust storms – and that dust can come to rest on the solar panels. In short, try building an electric vehicle (EV) for the worst possible scenario and make is 100 percent sustainable. It's not easy, but the lessons learned have Earthbound applications.
Jónsson said the inspiration for the program came in 1999, when NASA was in a very difficult position. The agency had just lost two Mars missions and future funding was not guaranteed. Instead of retreating, NASA decided to go for the gold and prove that they could successfully send rovers to the planet. It was because the organization's virtual back was totally and completely against the wall that people were willing to try new things. Read on after the jump to see what those were, and how they'll come in handy in the near future.
While the missions were being developed, the engineers realized they didn't have much – no, any – room for error. Everything needed to be absolutely perfect. They knew that a Martian day is 24 hours and 40 minutes long, and used as a starting point to plan out how to collect enough energy during each day in order to conduct the tasks the scientists wanted to do and give enough time to charge the batteries so they could power the rover through the dark Martian night. One important details: the batteries needed to be able to turn the rover on again the next day. If they ever used too much energy and left the rover without enough energy to power up, that would be the end of the mission. Thus, the first hard and fast rule was that batteries were never allowed to go below 40 percent charge. Ever. No matter what the researchers wanted.
Jónsson said that the scientists wanted to do everything – they wanted to be able to move the rover all the time, to take lots of high-resolution images, to scrape every rock they could and see what's underneath the surface – but somehow they needed to learn that battery status and sunlight forecast were true limiting factors, and that they would need to respect them. The key to the solution was intelligent software.
Jónsson and his colleagues developed an AI system that monitored pretty much every detail of the rovers' power levels. But it wasn't enough to have the computer tell the researchers what the best course of action was. The scientists are people, and people like to have options and to be in control. So, how did the system convince the scientists to change their desired behavior? By giving them the options they wanted and making them think they were in control.
In testing, there were three overall ways the scientists could control the rovers. First, they could just "trust the system" to do everything. The scientists would tell the rover to move three meters to the north, for example, and the rover would do so when the AI said it would have enough power to do so. Second, the AI could listen to what the researchers wanted and then offer piecemeal suggestions many times along the way so the work would get done in the most efficient way possible. Third, the AI would let the humans put in their controls and the rover would follow those instructions – as long as the one rule that they couldn't kill the robot was followed (want to drive all the way over there? sorry. maybe tomorrow). Over time, the piecemeal suggestion method turned out to be the most acceptable. Interestingly, and here's where we get into some day-to-day, Earthbound benefits, once the researchers developed a little bit of trust in the system and got to see the continued benefits of following the AI's suggestions, they were more and more likely to keep trusting the system with to make more and more decisions.
How well has this worked out on Mars? Well, the plan had been to run the rovers on Mars for 90 days. Instead, the missions have been going for six and a half years (and counting). The original plan plan called for one rover to travel 600 meters and for both to cover a full kilometer between them. By now, the two rovers have traveled over 30 km, combined (sadly, one of the rovers cannot move any more and is being used as a stationary lab).
So, how do these lessons help electric vehicle engineers on Earth? By realizing that the scientists and EV drivers are all humans, with similar wants. EV drivers want to be able to driver everywhere they want to go, want to be able to change their mind on the way home and so forth. Just think of the scientists as EV consumers and transfer the lessons over. Here are the lessons in bullet form:
- Mobility can be totally sustainable when is has to be (not that everyone would be happy with this, but it can be. Just imagine if we all went back to walking everywhere).
- You can't just tell people to not drive (in the Mars case, it didn't work to just tell the scientists they couldn't poke rocks and take pictures).
- A well-developed AI can change behavior.
- Most importantly, AI systems can a.) achieve more with less and b.) change behavior without affecting the happiness.
Jónsson's research showed us that we have to look at the whole picture – power sources, objectives, behavior, external circumstances – to understand how to make a system truly sustainable. You can't just focus on hardware (the cars, powertrains, etc.) or just on social factors (telling people to change their habits). The system has to change people's behavior and there need to be incentives that make people happy when they change behavior. Jónsson's work also shows that humans want to be in control, and think they make the smartest decisions. Adopting AI-based electric car management systems to this reality – by modeling human behavior, optimizing against expectations and making "suggestions" part of the process – can help us conserve energy in out EVs. The reality is that people are, to some degree, quite predictable and a good AI system can take advantage of this.
(NASA's videos say they're embeddable, but people are having trouble watching them, so here's a link to NASA's website where they are hosted.)
Our travel and lodging for this coverage were provided by the event organizers.