Fachbereich Informatik

Shadow-Hand Calibration Tool

Master-Thesis at group TAMS


The Shadow Dexterous Robot Hand is an advanced robotic hand that closely matches the size of a male human hand. With four joints per finger and five joints for the thumb, and an actuated wrist, it can mimic a large subset of human grasps and manipulation motions.

However, performance of the hand is often limited by the lack of automatic calibration, as actual finger positions often differ from the commanded positions. This is especially critical if fingers collide, as this results in high forces on the tendons (which can break) and the motors (which overheat quickly). Also, the tactile sensors on the fingertips are very sensitive and can be overloaded and damaged if finger positions are wrong.

The current calibration approach uses a complicated manual process, where the fingers are aligned with mechanical jigs one by one, and the corresponding sensor readings are then entered into configuration files. Despite their complex and nonlinear behaviour, there is no calibration procedure for the BioTAC tactile sensors at all.

Automatic Calibration Tool: Hardware

The core idea of the proposed thesis is to develop a single (or a set of a two or three) sensing object to be grasped by the Shadow Hand during an automated calibration sequence. The tool will be shaped so that the fingers reach and touch sensors (switches, light interrupters, force sensors) during the calibration process, allowing the software to record precise finger motions and also finger forces.

The whole process must be designed to be safe for the hand, even when starting from uncalibrated initial finger positions which might have collisions with other fingers (or the tool). One possible approach is to start the hand in a limited low-power mode for initial rough calibration, and to repeat the process with higher power and precision once the initial state of the hand is known.

After basic calibration of the hand geometry, the next step in the automatic calibration sequence targets to record the tactile sensor outputs, pressing against the force sensors in the calibration tool to determine zero-bias and gains. For the BioTAC sensors it would also be interesting to calibrate their reaction to contacts at different parts of the sensor, and the reaction to shearing-forces applied on the center of the sensor.


The precise scope of the software implementation is to be discussed, but the software must be integrated into the ROS framework and the EtherCAT-based real-time driver for the Shadow Hand.

Thesis Goals: