Robot components · engineering guide

Robot Cameras: Depth, Stereo and Event Vision Sensors

Compare verified depth, stereo, time-of-flight and event-camera products used in robotic perception.

Engineering analysis and field questions

Robot cameras are selected by the measurements a task requires. Stereo depth, time-of-flight depth and event sensing have different failure conditions, computing demands and calibration procedures. A camera capable of producing a 3D image may still provide unusable measurements on a reflective object or a low-texture surface.

Should a bin-picking robot use stereo or time-of-flight depth?

A stereo camera uses images from two viewpoints to estimate displacement and distance. It needs enough identifiable detail in the image pair; plain white packaging, glare and repeated patterns can frustrate correspondence. An active time-of-flight camera estimates range from reflected light, which introduces different errors on shiny metal, transparent materials and surfaces producing several reflected paths. Before selecting a camera, define the object dimensions, working distance, minimum detectable feature and available lighting.

A suitable comparison uses one measured calibration object at several positions in the actual bin. Record absolute depth error, the percentage of pixels with valid range and the repeatability of the robot's estimated grasp pose. For an industrial picker, accepted placements after the full robot motion matter as much as the camera's raw depth map. A camera that measures an empty bin accurately can still fail once objects overlap.

When is event vision useful for robot motion?

Event cameras send information when brightness changes at individual pixels, with timestamps for those events. They are useful to study rapid motion under appropriate lighting, but they do not continuously provide the same complete image as a conventional RGB camera. An object sitting still can produce few events. A manipulator that needs both object identity and fast motion cues may combine an event sensor with another camera.

Evaluate a moving high-contrast target at several speeds, including a stop. Keep exposure, room lighting and timestamps documented. LED flicker can generate event traffic unrelated to the object; a slow host or USB link can then fall behind on short bursts. The useful comparison includes end-to-end detection delay and missed detections, with the processing hardware named.

What calibration is needed before the first robot pick?

Identify the camera frame, robot base frame and tool frame. Calibrate camera intrinsics and measure the transform between camera and robot with an appropriate target. Repeat the calibration check after any mechanical adjustment to the lens, camera bracket or gripper. A two-millimeter mounting shift can be more damaging to a precision pick than a small improvement in depth-sensor specifications.

Document the camera's depth range, frame rate at the chosen resolution, transport interface and host processing requirements. For a mobile robot, add vibration and temperature tests. Camera documentation supplies the reference configuration; your own test establishes accuracy in the robot's actual environment.

Original technical documentation

Products and specifications to verify

These records refer to different configurations. Open the individual product to inspect its interfaces, stated specifications and official source before comparing performance.

RealSense Depth Camera D455

RealSense, Inc. · Stereo depth camera

Stereo depth camera. Global-shutter depth sensors.

USB

Stereolabs ZED 2i

Stereolabs · Stereo camera

Stereo depth and RGB imaging. Integrated IMU.

USB 3

Basler blaze

Basler · Time-of-flight depth camera

Time-of-flight 3D camera family. Industrial housing options.

GigE

Prophesee EVK4

Prophesee · Event camera evaluation kit

Event-based vision sensor evaluation kit. High temporal resolution.

USB

All robot components