From data
To meaning

CELI won twice at the Reply’s AWS DeepRacer Competition in just one week, taking two podiums.

We participated to two competitions involving autonomous cars: the first in Turin, on the 27th of June (during the Italian Tech Week) and the other one in Milan, on the 3rd of July. We obtained great results and we had the chance to meet a lot of Reinforcement Learning (RL) enthusiasts!

CELI is a company with twenty years’ experience in Natural Language Processing: during our activity, we built a team of Machine Learning experts. Three young members of this team challenged themselves in this different context, designing the algorithms that made us win the two competitions.

What about the race? The competition formula consists in letting some autonomous cars challenge each other on a real track. These cars – 1/18th scale race cars – are trained through Reinforcement Learning models. Contestants have 4 minutes to perform a complete lap of the track in the shortest time possible with their RL-trained model; every self-driving car can’t go off track more than 3 times – otherwise, time on the lap will be cancelled.

CELI’s team won and took the entire podium of Turin’s race and finished third in Milan!

Autonomous cars are vehicles capable of moving exclusively through signals transmitted by sensors – such as radars, sonars and GPS systems – positioned on it, without any human-generated help or input. A specific algorithm interprets information coming from the autonomous car, identifying the best path to take and avoiding possible obstacles.

auto guida autonoma addestrata RL

Autonomous cars’ models are trained through Reinforcement Learning (RL). RL is a particular area of Machine Learning whose main purpose is to design software with decision-making skills in an environment in which the agent stands – received through sensors on the vehicle.

The main goal of the agent is to move in the environment to maximize a function, known as reward function, able to reward or punish the agent’s actions as long as they’re right or wrong.

modello Reinforcement Learning

Together with Supervised Learning and Unsupervised Learning, Reinforcement Learning is one of the three basic paradigms of Machine Learning.

Do you want to know more about our work in Natural Language Processing, Machine Learning and Reinforcement Learning? Contact us at