Photo: MIT
Waiting at a red light is hardly anyone’s dream, but this is inevitable in big cities. With the advent of autonomous vehicles though, idling at red lights may become a thing of the past. The researchers at MIT demonstrated a machine-learning approach to controlling a fleet of autonomous vehicles.
We’ve seen this so many times when driving in the cities. People like a good feel of acceleration without considering the traffic in front of them. Then they slam the brakes and wait, idling at the red light. Do they learn anything in the process? No, they will repeat the same procedure over and over again. That’s what powerful engines are for, right?
The waiting time is not the only thing unpleasant with the signalized intersections. Vehicles consume fuel and emit greenhouse gases while waiting for the light to change. Even with the start&stop systems of today and electric cars, energy is wasted on air conditioning and ancillary equipment. Researchers have tried to find solutions to this nuisance. The first traffic sensing systems appeared to adjust red light timings depending on the number of vehicles waiting to cross the intersection.
But more can be done, and MIT researchers are all involved. Human drivers might find it hard to time the red light and adjust the speed accordingly. But this could be achieved consistently by an autonomous vehicle that uses artificial intelligence to control the speed. And with autonomous vehicles slated to take over the streets soon, this could be an exciting proposition.
In a new study, MIT researchers used artificial intelligence to control a fleet of autonomous vehicles as they approach and travel through a signalized intersection to keep traffic flowing smoothly. This has reduced fuel consumption and emissions while improving average vehicle speed. The best results are achieved when all the vehicles on the road are autonomous. Still, it can also work when only 25% use the control algorithm.
The scientists used deep reinforcement learning, where the control algorithm learns to make a sequence of decisions. Thus, the algorithm leverages assumptions learned by a neural network to find shortcuts to good sequences, even if there are billions of possibilities. But the researchers want the system to learn a strategy that reduces fuel consumption and limits the impact on travel time at the same time, which can be conflicting.
“To reduce travel time, we want the car to go fast, but to reduce emissions, we want the car to slow down or not move at all. Those competing rewards can be very confusing to the learning agent,” says senior author Cathy Wu.
During the testing phase, the team at MIT found out that their algorithm didn’t create any stop-and-go traffic, at least when only one intersection was modeled. If every vehicle on the road is autonomous, their control system can reduce fuel consumption by 18 percent and carbon dioxide emissions by 25 percent while boosting travel speeds by 20 percent. Even when the algorithm controls only 25% of the vehicles, this can offer 50 percent of the fuel and emissions reduction benefits.
As we’ve said, this study focused on a single-intersection algorithm. Still, in real life, things get complicated when intersections cascade. The researchers will further extend their analysis to include multi-intersection scenarios and different types of intersections. The work is still in its early stages, but Wu thinks this approach could be implemented near term.
The waiting time is not the only thing unpleasant with the signalized intersections. Vehicles consume fuel and emit greenhouse gases while waiting for the light to change. Even with the start&stop systems of today and electric cars, energy is wasted on air conditioning and ancillary equipment. Researchers have tried to find solutions to this nuisance. The first traffic sensing systems appeared to adjust red light timings depending on the number of vehicles waiting to cross the intersection.
But more can be done, and MIT researchers are all involved. Human drivers might find it hard to time the red light and adjust the speed accordingly. But this could be achieved consistently by an autonomous vehicle that uses artificial intelligence to control the speed. And with autonomous vehicles slated to take over the streets soon, this could be an exciting proposition.
In a new study, MIT researchers used artificial intelligence to control a fleet of autonomous vehicles as they approach and travel through a signalized intersection to keep traffic flowing smoothly. This has reduced fuel consumption and emissions while improving average vehicle speed. The best results are achieved when all the vehicles on the road are autonomous. Still, it can also work when only 25% use the control algorithm.
The scientists used deep reinforcement learning, where the control algorithm learns to make a sequence of decisions. Thus, the algorithm leverages assumptions learned by a neural network to find shortcuts to good sequences, even if there are billions of possibilities. But the researchers want the system to learn a strategy that reduces fuel consumption and limits the impact on travel time at the same time, which can be conflicting.
“To reduce travel time, we want the car to go fast, but to reduce emissions, we want the car to slow down or not move at all. Those competing rewards can be very confusing to the learning agent,” says senior author Cathy Wu.
During the testing phase, the team at MIT found out that their algorithm didn’t create any stop-and-go traffic, at least when only one intersection was modeled. If every vehicle on the road is autonomous, their control system can reduce fuel consumption by 18 percent and carbon dioxide emissions by 25 percent while boosting travel speeds by 20 percent. Even when the algorithm controls only 25% of the vehicles, this can offer 50 percent of the fuel and emissions reduction benefits.
As we’ve said, this study focused on a single-intersection algorithm. Still, in real life, things get complicated when intersections cascade. The researchers will further extend their analysis to include multi-intersection scenarios and different types of intersections. The work is still in its early stages, but Wu thinks this approach could be implemented near term.
Google News