ViseJoint is a virtual sensor based on computer vision software that measures angular positions of joints (virtual encoder). All axes of a robot arm are measured simply by looking at them with cameras. Positioning and control of robot arms works purely camera-based, even a robot arm without any hardware encoders is possible!
ViseJoint relies on one or several standard cameras next to the robot arm. Smart service robots already use cameras for navigation, pose estimation or object/face detection. Now, those existing cameras are co-used for angular sensing! All data derived from the camera is naturally coherent, ensuring minimal relative positioning errors.
ViseJoint enables sensorless robot arms, relying only on an external camera. Moreover, with external sensing, the true position of the arm is always known. Gear backlash, bending of the links or imprecise mounting no longer disturb positioning accuracy. Thus, camera-based sensing greatly simplifies the links, gears, sensors, electronics and wiring of a robot arm. Novel home applications or ultra low-cost robots become possible!
Details An external, nearby or “eye-in-hand” camera observes the robot. This remote sensing approach avoids errors by decalibration and ensures results coherent with the outside world. Also, the camera is co-used for other tasks like object recognition and bin picking.
Details Sensor data are calculated virtually by a computer vision software, observing the robot with a camera. Electronic components and cabling required by standard hardware sensors are replaced with a software solution!
Details Camera-based sensing shifts hardware functionality to software, reducing the number of electronic sensors and their wiring. But there is more: External sensing captures the true state of the robot, including backlash and bending. Therefore, simplified and lighter mechanical components like gears and links can be used!
Flexible software library for realtime joint angle measurement from cameras and images.Contact us
Customized interface or ROS (ROS1/ROS2) integration including MoveIt & interfaces for cameras, joint statesContact us
Full software, robot model, electronics and hardware for Igus desktop arm & miniature armContact us
From concept to product: We support you with integration of camera-based arms on your robotContact us
Realtime angle measurement (up to 7 dof) using cameras and computer vision software. Sensorless motion control for smart robot arms on service robots and in mobile or home applications.
A concept for a next generation vacuum robot: The mobile platform features a sensorless robot arm solely controlled by cameras, which obtain joint positions from visual markers on the arm.
Interview and demo of a robot arm fully controlled by camera-based angular sensors. This demo on Hanover Fair shows a pick and place application on low-cost Igus hardware. By automotiveIT.
Fully camera-based sensing and control of a "sensorless" robot arm. This proof-of-concept uses Igus gears and has been presented on A3 business forum.
|ViseJoint camera-based angular sensors(1)|
|Camera setup||Many options, e.g. 2-3 cameras in workspace next to robot|
|Camera type||Standard grayscale or RGB global shutter cameras|
|Relative positioning accuracy||Up to 1 mm (2)|
|Measurement rate||Typ. 60 Hz; 20 Hz to 120 Hz feasible (camera framerate)|
|Number of axes||Unlimited; typ. 5 to 7 joints|
|Max. motion speed||1.5 m/s|
|Marker placement||Flexible; typ. on base, end-effector, middle link|
|Interfaces and hardware|
|Camera interface||USB3Vision, GigE Vision, UVC, any ROS camera|
|Software interface||ROS (standardized incl. MoveIt, ROS Control); Customized|
|Software package||Docker containers on Windows/Linux; Ubuntu package|
|Hardware requirements||Standard PC with i5/i7 CPU; Various ARM platforms|
Camera-based sensing can be configured in many ways, adapted to your needs.
For instance, a higher camera resolution and a zoom lens improve accuracy.
Above, you find typical values with mid-range industrial cameras.
(2): The limiting factor is pose estimation of the object. Robot pose estimation is typically more accurate.