Mimir System

Mimir is an intelligent self-driving system based on convolutional and recurrent neural networks.
We put the rejection of HD-maps at the heart of our technology and plan to use lidar only to refine the data in Sensor Fusion. In the future we plan to abandon it. Therefore, we are focused on computer vision and now we're doing Monocular 3d Object Detection.
See more cases on Youtube
Objects, pedestrians, traffic lights, signs, lanes

Computer Vision


We use IMU and GNSS sensors to help a vehicle determine its current location.
Satellite navigation (GNSS) in urban environments often causes interference that is critically dangerous for an unmanned vehicle to rely on, therefore, we base the localization data on the IMU sensor and telemetry from the vehicle itself.


The Planner module is responsible for planning the route, passing intersections and difficult sections (for example, exits).
The Control module, which uses the nonlinear MPC method, controls the car at a specific moment by setting the steering angle, acceleration, and braking.


We develop self-driving software operating under any weather conditions. We can find lanes on damaged, snowy roads and roads without road markings.
High-precision maps are not used to navigate and drive the shuttle. This reduces shuttle operation cost and increases its market value.
Our approach includes reinforcement learning which evolves our vehicles while driving.
See more cases on Youtube
Road lanes

Winter night