Home          Experience          Research          Teaching          Publications

Projects

• Machine Learning
• Robotics
• Computer Vision

Robotics

---

Novel Interaction Strategies for Learning from Teleoperation (PR2 Robot)

The field of robot Learning from Demonstration (LfD) makes use of several input modalities for demonstrations (teleoperation, kinesthetic teaching, marker- and vision-based motion tracking). In this project we present two experiments aimed at identifying and overcoming challenges associated with using teleoperation as an input modality for LfD. Our first experiment compares kinesthetic teaching and teleoperation and highlights some inherent problems associated with teleoperation; specifically uncomfortable user interactions and inaccurate robot demonstrations. Our second experiment is focused on overcoming these problems and designing the teleoperation interaction to be more suitable for LfD. In previous work we have proposed a novel demonstration strategy using the concept of keyframes, where demonstrations are in the form of a discrete set of robot configurations. Keyframes can be naturally combined with continuous trajectory demonstrations to generate a hybrid strategy. We perform user studies to evaluate each of these demonstration strategies individually and show that keyframes are intuitive to the users and are particularly useful in providing noise-free demonstrations. We find that users prefer the hybrid strategy best for demonstrating tasks to a robot by teleoperation.

Compiled video here.
Download the paper here.

---

Robot Learning from Demonstration - Teaching Strategies (PR2 Robot)

We are interested in developing learning from demonstration systems that are suitable to be used by everyday people. We compare two interaction methods, kinesthetic teaching and teleoperation, for the users to show successful demonstrations of a skill. In the former, the user physically guides the robot and in the latter the user controls the robot with a haptic device. We evaluate our results using skill dependent quantitative measures, timing information and survey questions. We find that kinesthetic teaching is faster in terms of giving a single demonstration and the demonstrations are more successful. However, the learned skill does not perform better as expected. The survey results show that users think kinesthetic teaching is easier and more accurate and an open-ended question suggests that people would prefer kinesthetic teaching over teleoperation for everyday skills.

Compiled video here.
Download the report here.

---

Humanoid Robot Learning using Gaussian Mixture Models (Nao Robot)

A system was developed for robot behavior acquisition using kinesthetic demonstrations. It enables a humanoid robot to imitate constrained reaching gestures directed towards a target using a learning algorithm based on Gaussian Mixture Regression. The imitation trajectory can be reshaped in order to satisfy the constraints of the task and it can adapt to changes in the initial conditions and to target displacements occurring during the movement execution. The potential of this method was evaluated using experiments involving Aldebaran's Nao humanoid robot and Fawkes, an open source robot software by the KBSG at RWTH University.

Compiled video here.
Download the report here.

Update -

When teaching the Nao complex behaviors that involved using many actuators, the Gaussian Mixture Regression (GMR) model was found to be very slow and not suitable for online applications. In order to overcome this, the Approximate Nearest Neighbour Search (ANNS) algorithm was tested as a regression model in the Nao Simulator (Webots). The time complexity and the prediction errors were analyzed and a graph of their performance against the GMR can be found here.