We are building a great team to solve great challenges

Classical SLAM and visual odometry algorithms need manual parameter tuning for specific environments. Deep learning based VO algorithms can extract these parameters from data during training and are more suitable for robust positioning in space. Adding inertial measurement unit and multiple video channels inputs also contribute to safer navigation.

Any physical agent moving in 3D space has to do it in safe and reliable manner. To make it happen we are using classical path planning algorithms (like A*) as well as deep learning approach like TrailNet DNN. Our robots are capable of navigating not only in known map environment but also in search and discover mode using our own exploration algorithms.

Modern robotic devices use different kind of physical sensors to map their surroundings: ultrasonic, laser, radar or vision based. All these data streams need to be fused in a form of global point cloud where old data is updated and dynamic objects are detected. We are in a constant search for best sensors available on the market and ways to integrate it into our autopilot system.

To enable safe and useful collaboration between humans and robotic agent, latter need to be aware of presence of humans and other objects around them. Being edge intelligent device our AI navigation module is capable of real time on-board object recognition and telling between humans (that are to be treated with respect and care) and other obstacles. Also object recognition is vital for human like navigation and determining waypoints in the presence of visual odometry positioning error.

Both classical and machine learning based navigation algorithms as well as AI capabilities require significant amount of computing resources. Putting all of them into space and power constrained robotic system is a great challenge. We are constantly exploring various kinds of modern computing hardware components, not only well known Nvidia Jetson/Xavier solutions but also emerging AI ASIC accelerators like Intel Movidius Myriad VPU and Google Coral devices.

Semi or fully autonomous robots would still require human control and supervision. Remote robot control interface is vital for commercial and personal usage of robot agent in real life cases. Thus we pay great attention to developing intuitive and user friendly web based interface that delivers comfort operation, full situation awareness and telepresense control from any remote part of the world.

Our AI navigation module integrates lots of hardware and software components. We are using ROS middleware to simplify data transfer and take best practices from open source robotic community. Sometimes we even write our own device drivers and web communication modules. Supporting clear software architecture to keep things simple and efficient is an important part of our every day work.

Working close to hardware requires low level hardware programming skills. We write real time and bare-metal application for micro-controllers and develop FPGA hardware circuits. If you have expertise in such field and want to grow into robotic applications get in touch to join our professional team.

Modern robotics applications include almost every aspect of electrical engineering: power, control, electronic, communications, instrumentation, computer and so on. All these subsystems should be designed to meet space, power and cooling constraints found on any robotic platform. Also edge AI computing requires low-latency and high throughput data transfer interfaces like PCI Express, CSI, USB 3.0 adding issues for PCB designers. All together it presents a exciting challenge for electrical engineers.

We are interested in mechanical engineers who are familiar with rapid design and prototyping for testing theories and concepts, ready for testing "in field". Awareness of the UAV concepts and participating in robotics challenges will be an advantage.