Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Electrical and Computer Engineering


Dr. Mehrdad R. Kermani


The field of human-robot interaction has garnered significant interest in the last decade. Every form of human-robot coexistence must guarantee the safety of the user. Safety in human-robot interaction is being vigorously studied, in areas such as collision avoidance, soft actuators, light-weight robots, computer vision techniques, soft tissue modeling, collision detection, etc. Despite the safety provisions, unwanted collisions can occur in case of system faults. In such cases, before post-collision strategies are triggered, it is imperative to effectively detect the collisions. Implementation of tactile sensors, vision systems, sonar and Lidar sensors, etc., allows for detection of collisions. However, due to the cost of such methods, more practical approaches are being investigated. A general goal remains to develop methods for fast detection of external contacts using minimal sensory information.

Availability of position data and command torques in manipulators permits development of observer-based techniques to measure external forces/torques. The presence of disturbances and inaccuracies in the model of the robot presents challenges in the efficacy of observers in the context of collision detection. The purpose of this thesis is to develop methods that reduce the effects of modeling inaccuracies in external force/torque estimation and increase the efficacy of collision detection. It is comprised of the following four parts:

1. The KUKA Light-Weight Robot IV+ is commonly employed for research purposes. The regressor matrix, minimal inertial parameters and the friction model of this robot are identified and presented in detail. To develop the model, relative weight analysis is employed for identification.

2. Modeling inaccuracies and robot state approximation errors are considered simultaneously to develop model-based time-varying thresholds for collision detection. A metric is formulated to compare trajectories realizing the same task in terms of their collision detection and external force/torque estimation capabilities. A method for determining optimal trajectories with regards to accurate external force/torque estimation is also developed.

3. The effects of velocity on external force/torque estimation errors are studied with and without the use of joint force/torque sensors. Velocity-based thresholds are developed and implemented to improve collision detection. The results are compared with the collision detection module integrated in the KUKA Light-Weight Robot IV+.

4. An alternative joint-by-joint heuristic method is proposed to identify the effects of modeling inaccuracies on external force/torque estimation. Time-varying collision detection thresholds associated with the heuristic method are developed and compared with constant thresholds.

In this work, the KUKA Light-Weight Robot IV+ is used for obtaining the experimental results. This robot is controlled via the Fast Research Interface and Visual C++ 2008. The experimental results confirm the efficacy of the proposed methodologies.