Running Kudan’s VSLAM on the latest Qualcomm® Robotics RB5 platform
Earlier in the summer, Qualcomm announced its latest robotics development kit, the “Qualcomm® Robotics RB5 Platform”, and we posted about our partnership to enable robotics applications using Kudan’s SLAM solution.
We’ve been busy playing around with the RB5 board to see how well our visual SLAM performs, and wanted to share some of the results. We ran it through the EuRoC dataset (a well known standard dataset to test visual SLAM) as a warmup, but also built a small course in our office to simulate some real life indoor environment that an industrial AMR or commercial robot could come across.
With the EuRoC evaluation data, our VSLAM solution ran at 20 fps (live frame rate) and succeeded in maintaining tracking at 100% without losing its position at any given point in time. A bit boring… 😉
Now for the fun part, we created a human powered “robot” by mounting an Intel RealSense D455 camera, connected to an RB5 board on a trolley, and ran it through the course created at our office, eventually drawing a loop.
We initially ran the “robot” through the course to create a map in advance, and then we tested re-localization on that map. We had some white wall challenges in the path that would freak out other feature-based visual SLAM solutions, but as you can see, we were able to return to the original position without losing tracking.
We were pleasantly surprised by the ease at which we could pair up the Intel RealSense camera and the Qualcomm RB5 board, and get Kudan’s SLAM solution up and running. With the ready availability of both camera and development platform, it’s no surprise that many robotics players have opted to use this combination. The stability of the RB5 platform and RealSense combination should allow developers to improve overall performance through software updates without having to further tweak the camera sensors.
Please feel free to contact us for more detailed information!
【For your reference】