Projects and Publications

Alexander Werner

My interest in robotics are in two fields:

understanding control of elastic actuation principles and using system dynamics to enhance a robots capabilities
create, maintain and explore the capabilities of complex robotic systems

I work at RoboHub of the University of Waterloo. You can contact me at e-mail address

VR/Multi-view mobile teleoperation with haptic feedback

To understand how effective a user can be in teleoperating a manipulator with just a mobile phone, I have created a prototype. In this demonstration multiple pointclouds from RGB-D cameras are compressed, streamed over WebRTC to a mobile phone, and displayed in a VR environment which the user can navigate freely. By holding a latch button the motions of the phone are replicated by the robot, allowing the user to control the full 6D motion of the robots tip. To increase immersion, any measured external forces estimated by the robot are conveyed to the user by hapic and audio signals. Haptic and audio signalling are particularly interesting as they allow low-latency feedback in contrast to any visual feedback.

Workshop on the Talos humanoid robot

Series elastic robot walking over obstacles

Robust walking on the series elastic robot C-Runner. This work combines the use of optimization generated trajectories for walking on series-elastic robots with the passivity based balancing control approach used in other projects. Due to the mechanical robustness of the system, unexpected impacts can be endured and the control scheme allows to robot to tackle obstacles of a height of 25mm and more. Work done together with Henze B, Loeffl L, Leyendecker S, Ott C. Published on Humanoids 2017.

Humanoid Robots in Aircraft Manufacturing

In this demonstration for Project Comanoid (a European Horizon H2020 Project) Toro is using visual servoing to walk up to the table and bring himself into a suitable position for grasping. The pink object (cable mounting bracket) is located at a known position relative to the April tag fixed to the table surface. This location is used to to plan a path which is given to a Cartesian impedance controller controlling the hand motion to grasp the bracket. Toro then turns around to walk up to the airplane part. Arrived there a hand contact is added to extend the reach of the robot to the target location marked by the red square. In the following part the center of mass of the robot is moved outside of the support polygon given by just the feet, making the hand contact a requirement. The robot attaches the (magnetic) bracket to the airplane and returns to bipedal contact. The whole process is fully autonomous. This video is unedited and at original speed. Work done together with Henze B, Porges O, Englsberger, Garofallo G, J, Roa MA and Ott C.

Enhancing Series Elastic Actuators with Damping

Series elastic actuators with soft springs can provide interesting performance gains over rigid actuation. A drawback is that soft springs also reduce the available bandwidth for controlling the link-side position or torque. This can be remedied by adding a damper in parallel to the elastic element. Work done together with Kim MJ and Loeffl F.

ICRA 2017 (accepted)

submitted to IROS 2017

Test setup for new spring damper component.

C-Runner: Dynamic Walking Experiments

Published on IROS 2017. Work of my master thesis student Wojciech Turlej. Compliant(SEA) Biped C-Runner is walking using optimized trajectories. The model used contains the full link side dynamics and can use arbitrary contact points. Additionally, the motor side limits are respected. The resulting trajectory is the executed on the robot using motor-side position control, the human in the loop stabilizes the base (estimated human force <20 Newton). Conference Slides

C-Runner: Concept and Design

Multi-Contact Planning and Control

Published at IEEE/RSJ Intl. Conference on Intelligent Robots and Systems (IROS) 2016: Multi-Contact Planning and Control for a Torque-Controlled Humanoid Robot - Werner A, Henze B, Rodriguez D, Gabaret J, Porges O and Roa MA. The master thesis of Rodriguez D was co-supervised by me. Toro is controlled using the Multi-Contact Balancer previously published. The task at hand getting up the stairs is formulated as a sequence of contact pairs. From this sequence, intermediate configurations are synthesized using inverse kinematics. A modified Rapidly-exploring Random Tree algorithm which operates on the manifold of the contact constraint is applied to generate a feasible path. Planning time for this trajectory is approximately ten minutes. The configuration space path is then heuristically converted into a trajectory. The paper also features automatic generation of contact points from RGBD data. The algorithm heuristically explores feasible steps leading from the current contact set towards the goal. The approach presented is applied to foot contacts only.

C-Runner: First Walking Experiments

Picking Objects from a Table

Unpublished: Work with Porges O. Toro walks up to the table using visual servoing based on the April tag mounted on the table. The blue cubes are then located based on RGBD data, from the perceived position one is selected. Even though the redundant part of the robot is used for grasping, not every hand position in SE3 is feasible. From the object position, a number of possible hand positions are computed and then verified against a reachability map. If the configuration is reachable, an inverse kinematic is queried to obtain the associated joint space configuration.

Generalization of Optimal Walking Trajectories

Published at IEEE/RSJ Intl. Conference on Intelligent Robots and Systems (IROS) 2015: Generalization of Optimal Motion Trajectories for Bipedal Walking - Werner A, Trautmann D, Lee D, Lampariello R. Trautmann D was supervised by me during his master thesis done in a collaboration between DLR and TUM. The figure on the right shows the difference local minima of a torque cost-function. This work compares suitable methods for generalization of a set of optimal trajectories and analyzes the errors in the process. Trajectories for bipedal walking are especially interesting with this respect as constraint violations can cause falling of the robot.

Control Applications of Toro

Robotics Library

The library implements the basic building blocks for robotics research:

Forward kinematics and associated Jacobians
Computation of matrix quantities for the robotic equation of motion, including time derivatives up to the second order
Recursive forward and inverse dynamics (including constraints)
Collision computation for primitives: Sphere, Capsule, Box based on quadratic optimization
Realistic non-linear contact model
Self-collision avoidance for torque controlled robots using potential fields
Limited inverse kinematics solver
OpenSceneGraph based visualization
URDF parser

The listed features is combined to a simulation environment. The library is implemented in Eigen, all algorithms are templated and can be used with automatic differentiation. Simulation and on-line dynamics are used in various systems throughout the institute. The library is also the basis for my motion planning works.

EMG based Tele-operation

Published in Frontiers in Neurorobotics: Stable myoelectric control of a hand prosthesis using non-linear incremental learning - Gijsberts A, Bohra R, Sierra Gonzalez D, Werner A, Nowak M, Caputo B, Roa MA and Castellini C.

The EMG signals measured at the forearm of the operator are used to drive the prosthetic hand mounted on the robot. A magnetic tracker measured the 6D pose of the operators lower arm which is replicated by the robot. A Cartesian impedance controller is used and the redundancy of the robot is leverage to obtain good tracking. Later a bi-manual variant of this experiment followed.

Trajectory Optimization for Series Elastic Joints (SEA)

Published at IEEE Conference on Decision and Control 2014: Trajectory Optimization for Walking Robots with Series Elastic Actuators - Werner A, Lampariello R and Ott C. This paper extend the previous work on optimization of trajectories for bipedal walking robots to series elastic robots (SEA). The approach uses a very small optimization problem as one spline function is used to parameterize the link side and the motor side. The video shows an optimized trajectory for an early design stage of C-Runner. The upper body tilt is due to the mass distribution which is not adequately displayed by the visuals.

Legged Humanoid Robot from DLR