First demo of our new aerial Visual-Inertial Positioning System (VIO+VPS): 40m positioning accuracy (CEP) in a 50 kilometer flight using a camera, consumer-grade IMU and a barometer! No GPS
GPS-free visual-inertial navigation in real time on Raspberry Pi 5 with a consumer-grade IMU and camera (and this is not even the slowest CPU this runs on!)
Easiest way to calibrate a camera on Linux. No ROS, no Docker, no CMake. Just @FFmpeg and Spectacular AI command line tool installable via Python pip:
sudo apt install ffmpeg
pip install spectacularAI[full]
sai-cli calibrate your-data.mkv
In the 3rd flight test, we basically duct-taped a smartphone to a Cessna to highlight that our GPS-free Visual-Inertial Positioning runs on consumer-grade sensors (32 m CEP positioning accuracy over 55 kilometers, only using phone camera, IMU and barometer)
New #drone demo! GPS-free navigation using VIO with two test payloads: one global shutter, one rolling shutter setup. < 1% accumulated drift in both cases.
We have now released Spectacular AI Mapping API: a low-level SLAM API for embedded devices. It can also be tested on OAK-D devices + PC. In addition, our C++ SDKs for OAK and RealSense devices are now publicly available for x86_64 (see 🧵 for links)
Mixed Reality with 135 lines of Python, and the new Spectacular AI SDK (on OAK-D). You could also replace the PyOpenGL rendering part with Unity or Unreal, but that would be less fun. With our SDK, you can also glTranslatef the #metaverse.
Another GPS-free aircraft navigation demo wth our new Visual-Inertial Positioning System: 77 km flight, 21 m CEP. If you are interested in early access of the solution, send us a message at spectacularai.com/#contact . More technical details in 🧵
Spectacular Rec app released for iOS and Android! Combined with a good GPU and Nerfstudio, our tools enable fast #gaussiansplatting reconstructions without uploading the data to a cloud or using COLMAP. Results can be viewed in the browser! Links in 🧵
Features of the new Spectacular AI SDK 1.38 include:
⭐ ONNX-based VPS acceleration: allows full VIO+VPS operation even on Cortex A53s
⭐ Built-in IMU-camera calibration Python tool, based on our fork of Basalt (separate post to follow)
For more details:
Gaussian Splatting on the Move updated!
⭐IMU is now optional: supports motion blur or rolling shutter compensation with plain video data
⭐Existing work comparison: Outperforms BAD-NeRF
⭐Rebased on latest Nerfstudio for improved accuracy
Our first public outdoor drone demo! Spectacular AI SDK can be used to track the position of a drone without GPS. Demonstrated here with our low-cost reference HW design (the round thing under the off-the-shelf DJI drone) and compared to the on-board GPS as the ground truth
The simplest way to improve VISLAM accuracy: buy a better IMU. We finally tested this in practice: Built a device with two IMUs (the cheap @InvenSense MPU6050 and the high-end @MurataEurope SCHA634), and tested which one is better when all other aspects of the data and are fixed