back
esr_

~/esr_technology/platforms/autonomous_driving

~/technology/platforms

Autonomous Driving
Exploring an entire software stack
for self-driving cars

/

Idea

/

Making a car drive autonomously is one of the most demanding challenges in artificial intelligence (AI).

Through the independent research project Autonomous Driving, the ESR Labs AI team aimed to develop a full software stack for self-driving cars. The goal was to understand how artificially intelligent systems work, down to their smallest components.

Scope

  • +

    Perception: Development of a perception system with different sensors, allowing the vehicle to make sense of the vast amounts of data from its environment. This data was used to develop mathematical representations of the environment and predict future movements and intentions.

  • +

    Localization and mapping: Multiple approaches were tested to enable a vehicle to determine its position and plan its path correctly.

  • +

    Decision-making: Use of reinforcement and rule-based methods coupled on a high level of abstraction to efficiently explore selected maneuvers. This results in a semantic (tactical) decision, which is then forwarded as simple numeric tasks to a geometric path planner.

  • +

    Drive control: Utilization of the Paravan drive-by-wire system that controls electromechanical actuators, allowing the vehicle to change lanes, turn, accelerate, and break. To test the system without actuators, a simple LED indicator was developed to signal intended actions by the car.

  • +

    Platform: Development of a proprietary autonomous driving software platform, allowing all the software components to be combined into one fully deployable system. It holds functionalities such as the playback feature that vastly speeds up the software development process.

Features

  • +

    Sensor calibration

  • +

    Stereo vision

  • +

    Semantic segmentation

  • +

    Object detection

  • +

    Road edge detection

  • +

    Map-matched localization

  • +

    Linear-environment model

  • +

    Reinforcement learning

  • +

    Real-time path planner

  • +

    LED indicator

  • +

    Drive PX2 computing platform

  • +

    Custom vehicle gateway

  • +

    Custom component framework

  • +

    Recording and playback capabilities

We use cookies to enable website functionality, understand the performance of our site and serve relevant content to you. More information: Privacy Policy and Cookie Policy.