english only
Student Projects propositions
If you are interested in one of the below projects, Semester Projects or Master Projects, please contact the first person of reference indicated in each description either by telephone, or by email, or by visiting us directly to the LASA offices.
Robot trajectory modification to avoid singularityDynamical system approach to learn reaching movement via human kinesthetic teaching is a powerful method for robotic manipulation task in Cartesian space. However, the existence of inverse-kinematics (IK) solution along the trajectory has received so far little attention. Especially, including orientation constraints in the trajectory planning drastically decrease the IK solution space, since robots usually limited by its degree of freedom. In this project, the student should explore several methods to avoid singularity by modifying the dynamical system trajectory planning method. The expected procedure to the student are: 1) to understand existing dynamical system trajectory planning method, GMR which is trained by SEDS, 2) to understand existing reachable space modeling method which shows where a robot's hand can reach, and 3) to explore several solutions.
|
||||||||||||||||||
Two Hands Grasping using iCub humanoid robot A humanoid robotA humanoid robot requires two hands to grasp a big object (such as soccer ball or a tea pot). The goal of this project is to implement two hands grasping by extending the existing code for one hand catching. The work will be divided as follows: 1) to understand existing graspable model of an object (where on the object can the robot place its fingers) and reachable model of a robot (what are the location in space which the robot's hand can reach), 2) to extend the graspable model of an object to two hands and 3) to implement the system using iCub simulator and real robot.
|
 |
|||||||||||||||||
Producing full body emotional postures on the iCub robotInteraction with a humanoid robot, raises a set of expectations from a novice user. Without knowing the limitations of the robot, people expect it to act "naturally", by giving proper responses in a timely manner. Providing proper gestures as feedback to the user is one way of making human-robot interaction sustainable. However these gestures should not interfere with other ongoing tasks in which the robot is involved. For example if the robot is involved in an object manipulation task, it will not be able to use its hands to produce gestures, but could easily use its head. At the end of the project it is expected to have a functional library for producing body gestures, that will not interfere with the functioning of the robot. If time allows the student may also contribute to the evaluation of the resulted gestures by running a pilot user-study.
|
 |
|||||||||||||||||
Object shape estimation with active touch sensing"We not only see but we look, we not only touch we feel" --JJ.Gibson A desirable ability for robots to work in unstructured environment is to be able to learn an image of the environment and of its components by "touching" it. In particular, to be able to determine the shape of an object is of fundamental importance for dextrous manipulation skills. In this project, the student will use very novel touch sensors mounted on a dextrous robot hand to actively collect data from unknown object surface and estimate the surface of the object iteratively during the data collection. The general outline is as follows: (1) understand non-linear regression techniques (such as Gaussian Processes Regression algorithm) (2)implement this algorithm with C++ to learn 3 dimensional tactile map.
|
 |
|||||||||||||||||
Classifying tactile gestures on the iCub robotIn the context of human-robot interaction, the sense of touch is particularly important as a lot of information can be conveyed through tactile sensing, and distinct emotions can be communicated more reliably than through facial expressions. The goal of this project is to classify tactile gestures with the aim of inferring user's intentionality. The iCub is a humanoid robot, the size of a 4 year old child, covered with tactile sensors on its arms, forearms, torso, fingertips and palms, that respond to variations in pressure. Classification of tactile gestures can be done based on the following parameters: touch behaviors (like hitting, shaking, petting etc.), duration, and qualitative measures (e.g. intensity, location, duration, velocity). At the end of the project it is expected to have a functional library for detecting and classifying tactile gestures. If time allows the student may also contribute to the evaluation of the resulted interface by running a pilot user-study.
|
 |
|||||||||||||||||
Identification of Inertial Parameters of Robotic ToolsIn most robotic applications, tools specialized for the task at hand are mounted on the robot end effector. Examples from industry are welding-guns, paintbrush etc. For accurate control of the system, it is necessary to model the inertial parameters of the tool. Computing these parameters from CAD models or similar is time-consuming, tedious and inaccurate. An alternative to this procedure is to infer the inertial parameters from measurements of robot trajectory and torque recordings. This project will consist of reviewing literature on identification of tools, and implement a control module for the KUKA LWR that can be used to automatically estimate the inertial parameters of a mounted tool.
|
||||||||||||||||||
Programming Robota to learn to play PingPongThis project consists in programming a little game that people could play with Robota in which they can teach the robot to play pingpong. A video of Robota playing pingpong can be found here. The robot is displayed in the corridor of the MA building, 3rd floor. The application once developed will allow people to compete against each other to teach Robota the best shoots. Programming will be done in C/C++.
|
 |
|||||||||||||||||
Implementation of the whole body collision avoidance on the WAM arm (this project is assigned)Many robot tasks require reaching or placing an object while avoiding collision with other objects as well as the robot itself. The aim of this project is to take the existing code for the obstacle avoidance, that is only applicable to point-robots, and to extend it for the whole robot body collision avoidance. To obtain this, the student needs to find the closest point on the robot to the obstacle, and then drive it away (with the help of kinematic null-space). In this project, the following steps should be obtained: 1) To understand the existing C++ code and the approach, 2) to implement the closest point on the body to the obstacle, and 3) To drive the obtained point away from the obstacle.
|
 |
|||||||||||||||||
Implementation of gaze-arm coordination on the iCub robotThe aim of this project is to port the existing code for gaze-arm coordination, which was tested in the iCub's simulator, to the real robot. This work will tacle control of robot's gaze-arm coordination (with and without obstacle avoidance), visual object detection and grasping. A student will port already developed code on the real robot, test it and modify the code when needed, e.g. to deal with filtering of the noise which comes from sensors and cameras, to ensure robust behaviour in real-time, and to fine-tune control parameters.
|
 |
|||||||||||||||||
Teaching the Reeti robot to provide relevant information and adapt in a social contextImplement an affective module on the Reeti robot such as to be able to initiate an interaction and provide a rich communication in a social context (include various representations of affect, through facial displays, head gestures, gaze, variations in tonality). The robot should be able to use its sensing capabilities (vision, touch and sound) to infer people’s mood and adapt to their current attitude, by means mentioned above, while also giving useful information. Test the implemented module on a site guide scenario, where the robot is supposed to provide general site information during a pleasant interaction. The student may also contribute to evaluating the resulted interface by running a pilot user study for evaluating the robot’s behavior in real environment
|
 |
|||||||||||||||||
Incremental learning of force-based tasks using multimodal interactionIn the LASA lab, various methods for teaching robots how to do tasks by showing them have been developed. For example, a human teacher can teach a robot how it should move by teleoperating it while recording joint angles etc. Various machine learning techniques are then used to derive a task model that the robot can use to reproduce the task. Many manipulation tasks require that specific forces be applied on objects in the environment. Teaching this type of task is non-trivial, as it is much harder to demonstrate force than movement. In this project, various methods will be investigated to incrementally teach the robot a force profile for a task. For example the robot should be able to interpret vocal commands such as "push down", "harder" etc.
|
 |
|||||||||||||||||
Teaching a robot how to produce alphabet letters, similar to a child's handwriting (Project assigned)In this project we explore the concept of teaching handwriting to a child with the help of a robot and investigate various technical solutions to accomplish this. In the given approach, the interaction can be initiated by the child where he/she corrects mistakes made by the robot. Such a scenario would be built on the principle that learning by instruction can be more efficient for the child. The project involves developing a software that maps child's handwriting from an input device (such as a graphical tablet or inkling stylus)to robot motion. Child's handwriting is continuously evaluated and robot reproduction will include variations that will require the child to provide corrections and thus improve. If time allows the student may also contribute to the evaluation of the resulted interface by running a user-study. Project done in collaboration with CRAFT.
|
 |
|||||||||||||||||
Detection of transparent objectsThe aim of this project is to implement detection of transparent objects (glasses, transparent cups, etc.) in real-time for robotic application. Since transparent objects distort background and reflect the light, one shall exploit these cues in their detection. More specifically, we aim to exploit visually-discriminant features of transparent objects, found as latent components of SIFT image descriptor which is applied on a scalable image-spanning window. A classification algorithm is to be trained in order to recognize a transparent object in the image (for example, Support Vector Machines). At the end of the project, it is expected to have a functional C++ library for detection to be used for control of the iCub robot.
|
 |
|||||||||||||||||
Real-Time Trajectory planning for high-precision robot armThe work will entails the development of controller for Realtime Trajectory Planning to control a high-precision robot arm to help surgeons during in-vivo surgery. Work plan entails:
|
||||||||||||||||||