Our team with self-driving guru Sebastian Thrun in San Francisco receiving 1st prize in Udacity challenge
We successfully participated in competitions like
Udacity challenge (1st place in Challenge #3) and Didi challenge (7th place from 2 000 participants).
Sensor data fusion
We develop methods to integrate multiple data sources to produce more consistent, accurate and
useful world model to assist in self-driving capabilities.
One of the most important part in developing self-driving car is to simulate
as much driving as possible to reduce the cost of testing in real world. We are
currently using simulations in Gazebo and GTA V which enables us to develop new algorithms much faster and safer.
We cooperate with sensor suppliers to validate performance of the sensors
and measure KPI for particular testing scenarios.
OOnce the model of outside world is properly acquired, path planning
need to ensure the driving is save and lead to desired location.
How does Roboauto work?
You can see a lot of videos showing the functionality of Roboauto.
Check what Roboauto is capable of...
We use YOLO and Full Resolution Residual Networks (FRRN) segmentation to recognize the type of tracked objects.
Robot Operating System is the basic framework for our solutions.
Cameras, Radars and LiDARs
We use standard sensors to sense roads and objects near vehicle. We
also employ other available sensors, like GPS, IMU, odometry, car CAN
bus information, etc.
Lane detection is also implemented using deep neural networks and probabilistic algorithms.
AI - Neural networks
We are using deep neural networks widely to improve self-driving capabilities of
Roboauto. Every single drive or simulation helps to teach our neural networks and develop
even more to expect any kind of situation which can occur on road.
Probabilistic robotics pays important role in our algorithms. We widely use
Bayesian filters (e.g. particle filter, Kalman filter), probabilistic approach
(Markov chains, MCMC, Gibbs, Metropolis-Hastings) and its improvements for simultaneous
localization and mapping (SLAM), object based modeling surrounding environment and other problems in this field.
News from our Twitter
Interesting news from the world of self-steering car @roboauto
Roboauto is a self-driving startup located at Brno, Czech Republic.
Our team has started on 2007 on small car models and build first
full scale prototype on 2014. The team consist of 15 sw developers and technicians.