We develop 3D sensing software solution that could run on low-power computing platforms without using GPUs or hardware accelerators. Our software solution includes 3D reconstruction, object detection and path planning modules and can provide information on the robot/AGV/machine's surrounding to plan path and avoid collision. A core offering of RoboK is the construction of 3D point-clouds with critical object information efficiently on low-power computing platforms. The perception modules can therefore can be integrated on edge along with sensor modules to provide greater insights and intelligence without needing to be processed centrally first. The perception module will detect objects (e.g. workers, other machinery) and the information can be used to warn drivers and operators of potential collisions.Our solution can be deployed onto vehicles and machines like forklifts and provide valuable information for drivers to stay safe.

The software solution has been optimised for low-power...
We develop 3D sensing software solution that could run on low-power computing platforms without using GPUs or hardware accelerators. Our software solution includes 3D reconstruction, object detection and path planning modules and can provide information on the robot/AGV/machine's surrounding to plan path and avoid collision. A core offering of RoboK is the construction of 3D point-clouds with critical object information efficiently on low-power computing platforms. The perception modules can therefore can be integrated on edge along with sensor modules to provide greater insights and intelligence without needing to be processed centrally first. The perception module will detect objects (e.g. workers, other machinery) and the information can be used to warn drivers and operators of potential collisions.Our solution can be deployed onto vehicles and machines like forklifts and provide valuable information for drivers to stay safe.

The software solution has been optimised for low-power computing platform and enables robust and accurate perception for advanced robotics applications. 3D point-clouds are points in the 3D space that could be derived from laser scans, radar or 3D reconstruction based on images. One important benefit of working with 3D point cloud is that you can adjust the sparsity of the point clouds and working with sparse data lowers memory bandwidth requirement and power consumption. To ensure detection accuracy, RoboK has also developed efficient detection models specifically for sparse 3D point-clouds optimised for low-power computing platforms, which is different from applying deep learning models directly to raw 2D image data and then the class of the object, how far the object is and therefore the likely time to collision. Applying machine learning models on 3D point-clouds directly is more efficient as both object depth and class information can be obtained simultaneously. The efficient image processing and pixel-level sensor fusion algorithms can run on low-power CPUs such as Arm A53 cores without GPUs or hardware accelerators.
More information

Employees

Hao Zheng
Admin
Hao Zheng CEO An entrepreneurial economist and statistician who is passionate about making technology accessible.