Rovis.AI is our Artificial Intelligence software stack designed to enable real-time robotics applications based on Deep Neural Networks.
Our technology is based on three major pillars: connectivity, artificial intelligence and mechatronics. The connectivity pillar is represented by Rovis.DataChannel, which is our distributed computation system for interconnecting teams of robots, cloud systems and edge devices. Rovis.Vision is used to train and run deep neural networks in real-time for computer vision and perception. The low-level robotic control algorithms are implemented and run within the Rovis.Mechatronics software package.
Request a demo via Skynet.
Our approach to designing and deploying Artificial Intelligence applications is composed of a dual software systems:
Rovis.AI.workbench, which is a Python based software package used to design DNN architectures and train them on saved DataBlocks; it uses PyTorch, TensorFlow and ONNX (Open Neural Network Exchange) as low-level AI libraries.
Rovis.AI.inference, representing our C/C++ based optimized inference engine for trained DNNs running in real-time within the DataChannel.
The focus of our DNNs is on real-time inference for computer vision and perception. To achieve real-time capabilities, we use multi-tasking deep learning architectures which are sharing common backbones, while inference results are provided via independent network heads (e.g. object detection, scene segmentation, 3D reconstruction, features tracking).