The Ascent Approach


Deep Reinforcement Learning

Artificial Intelligence algorithms are not developed in the same way as traditional software. AI’s are trained in simulations through trial and error. Behaviour that produces better outcomes for the AI agent are reinforced while adverse choices are discouraged. This is called reinforcement learning.

This approach allows the AI to work out for itself which behaviours are most useful to master. We control the learning framework of the simulated environment and let the AI agent determine optimal behaviour models on its own.


Generative Models

The real world is complex and hard to model. Generative models learn how to mimic the real-world environment. They generate a set of circumstances relevant to the AI agents training. This is a more efficient way of teaching the AI compared to sending it out into the streets of a real city.

Generative models, in a word, generate environmental behaviour, situations and feedback for the simulated training setting. In particular, they can create rare events that are difficult to model or predict, thus preparing the AI for any situation, however unlikely, that might occur in the real world.


Transfer Learning

Traditionally AI’s have mastered only a single task. One of the challenges in developing machine intelligence comparable to human ability is to make sure that the AI is able to take advantage of past lessons as it learns new things.

Whether the AI is driving a car or a bus, in the city or on the highway, in a simulated environment or the real world, it learns the underlying principles applicable to all these tasks.

It all comes together in

Ascent Atlas

Ascent’s learning architecture, Atlas, is the skeleton on which the AI training simulations work. We are developing Atlas to enable more efficient and more intelligent training of AI’s, for a wide variety of tasks.

Incorporating our research on meta learning, we can significantly reduce the need for handcrafted code and labeled training data, as well as expensive real world testing.

To support Atlas, we are developing an intuitive 3D interface for advanced teleoperation and performance monitoring, which can be operated in both virtual and augmented reality environment.

Robotic Systems for the real world

Ascent’s research and development are aligned to bring cutting-edge AI to market. Even though our core competency is in AI algorithms, our modeling and research methodologies are strongly rooted in real world problems. We are an applied development team.

We have extensive knowledge in the sensing and guidance hardware required for the AI to interact with the physical world. This expertise is necessary when building comprehensive robotic systems for self driving cars.

Unique partnerships

Over the next few years, Ascent will engage with a select number of vehicle platforms and their manufacturers. We limit entering the market with a small number of products in order to maximize quality and capability but also to ensure commercial flexibility and speed.

Together with our industry partners, we will build amazing autonomous vehicles that are both intelligent and customer focused. Each partner platform will be customised to its own hardware, computation requirements, as well as interaction models and personality – every Ascent AI brand vehicle is a unique product.