Open main menu

Humanoid Robots Wiki β

Changes

Stereo Vision

19 bytes added, 13 May
no edit summary
Additionally, advanced users have discussed using the ZED2 camera for its superior baseline and resolution, and have compared various models for specific needs, such as indoor testing and 3D benchmarking with Kinect and Orbbec Astra cameras. The community also highlights the importance of considering off-the-shelf depth cameras that integrate stereo computation internally to save on development time and effort.
Users have shared their experiences with various depth-sensing technologies across different robotic platforms. For example, the Unitree robotics platforms and the mini-cheetah from MIT have incorporated RealSense cameras for mapping and environmental interaction. The broader field of view afforded by using multiple cameras, as seen with Spot which employs five cameras, is advantageous for comprehensive environmental awareness. However, for specific research applications where movement is predominantly forward, a single camera might suffice. The discussion Builders in the robotics community also highlights highlight the shift towards more sophisticated sensing technologies like solid-state lidars, though cost and weight remain significant considerations. For instance, the CEREBERUS team from the SubTChallenge noted that while lidars provide more efficient depth data than multiple depth cameras, their higher cost and weight could be limiting factors depending on the robotic platform.
This section reflects ongoing discussions and is open for further contributions and updates from the community.