Humans’ most intuitive skills are acquired through a natural process of observation, imitation, and practice. Replicating nature’s ways, we designed a system that observes its surroundings through real-time visual perception and learns motion planning by imitating the most skilled human drivers.

Imagry’s approach to spatial perception is to utilize specialized lightweight neural networks that can generate functional precise situational awareness models. These models can, in turn, be used by other modules within the system to create motion plans, interact with the environment and navigate within it.


Robust Functional Road

Imagry’s Minimap System produces dynamic road reconstructions translating visual input into local drivable maps.

Provided with the most natural and efficient sensory source of input – visual input – using regular and thermal cameras, the Minimap system utilizes a set of neural networks to achieve several goals:

  • Reconstruct a bird’s-eye view of the surrounding road topology, relying on a wealth of annotated road data covering all geometries, layouts and types
  • Detect and classify nearby objects
  • Position the classified objects accurately in 3D-space and predict their future trajectories, whether they are in line of sight or obscured by other objects

Human-Like AI

In a universe of infinite possible events, it is impossible to model reality using a finite, pre-prepared decision-tree. We designed a system that replaces the rule-based approach to decision making with a data-driven one. SpaceNet is Imagry’s spatial reasoning deep neural network that learns driving and creates motion-plans relying on real-time environment perception and learned-by-training driving behavior.

Using the DCNN approach, SpaceNet enables Imagry’s AI-driver to learn by mimicking human drivers’ reactions in a wide variety of driving scenarios, and then to extend the learned behavior and generalize it to situations that it hadn’t been specifically trained to handle. This generalization renders the explicit demonstration of every conceivable scenario unnecessary.


Adjustable to new roads and adaptable to unforeseen events, our system learns, internalizes and generalizes the driving function and rationale like a human driver.


Simplicity Is
the Key

Most Autonomous Vehicle systems

use multi-sensor fusion (cameras, Lidar and radar) and manually pre-annotated HD maps.

What it actually means is:
More components to process
(complex IT systems)
More processing time and significantly
higher risks of failures
Constant high bandwidth
communication demands
Risks of communication failures and
cyber-security attacks
Constant updating demandsExtensive ongoing investments and
labor force

Imagry’s AI-driver is designed to conquer possible systematic failures and master the driving task, with eyes on the road, mapping out every vehicle, pedestrian, traffic light and construction site in real time.Our minimalistic software stack is the product of lean efficient engineering allowing a significant decrease in IT related errors and unmatched cost efficiency

Independent of HD maps and multi-sensor fusion, Imagry’s system is unbound to external data sources or high bandwidth communication. As such the system eliminates all cmmunication and cyber security problems and saves expensive radar and Lidar costs as well as complex SLAM (Simultaneous Localization and Mapping) processes.


Testing Our Tech, Continuously
Raising the Challenge

Working by the ground rule that design and testing go hand in hand, we at Imagry insist not only on testing our tech, rather continuously raising the challenge. Since our technology operates through continuous, real-time, data-driven learning and practice, whenever our driver is on the road, its driving skillset is getting sharper. And this is how we constantly challenge, evaluate and improve the performance of our designs: we keep our system on the road; from closed courses and staged complex scenarios to real-time drives on highways and high-complexity urban environments in both day light and night time.