Computer vision module.

The new generation of robotic applications strongly depend on the ability to acquire, organize, and interpret surrounding visual data in real-time in order to stay aware of the environmental situation and act adequately. This ability is also known as “perception”.

Rubedo CVM is the stand-alone depth sensing and 3D perception system developed by Rubedo Sistemos. It employs passive stereo vision, which is the most affordable technology today for mobile or stationary robot navigation and object sensing.

One of the major driving forces for the 3D perception market is the increasing use of mobile robots and drones for multiple industrial and service applications. Without reliable environmental perception such robots are unable to detect obstacles and adaptively plan safe navigation paths. As a result, their operators must carry the burden of circumnavigation rather than concentrating on the direct operation tasks.

Service providers and equipment manufacturers often do not have the necessary skills and know-how to upgrade their existing products with computer vision-based AI required for context awareness. Instead, they would much rather use a standard solution. Currently in the market there is no truly affordable turn-key solution for quickly adding 3D perception and machine learning capabilities.

Rubedo CVM has been designed to bridge the above mentioned gap. It offers high-density, low-power, long-range optical depth sense and natural object recognition in an on-board package. Light and powerful, it can be applied to a wide range of use cases in industrial and service robotics domain.

Rubedo CVM provides high definition images, accurate measure of the environment depth, and can be trained to detect and track natural objects of interest in real-time.

Traditionally, LiDAR systems are used for reliable ranging and depth perception. However, the high cost of equipment, pilots and engineers kept the LiDAR technology at bay for most people and applications. Cutting edge LiDARs of today are still rather expensive devices, unable to output imaging modalities of a traditional video camera.

Two other popular methods for depth sensing – Time of Flight (ToF) and Structured Light Imaging (SLI) – are based on active emission of IR light, hence they require controlled lighting conditions and are mostly restricted for short range indoor use. Structure from Motion (SfM) technology – while attractive for its low equipment requirements – produces sparse depth maps and requires continuous motion in order to work, thus limiting application range. Finally, conventional stereo cameras have limited on-board image processing capabilities, shifting the burden of application-specific perception to the host system.

By contrast, Rubedo CVM has an integrated powerful NVIDIA Jetson TX1 GPU, which not only does on-board stereo matching, outputs pre-calculated disparities or dense point cloud to the user, but also can execute arbitrary image processing pipeline on behalf of a specific application (e.g. detect utility poles).