Context
This internship is part of a research project on haptic guidance. Haptic guidance provides force feedback to help an operator to remotely operate a robot thanks to a haptic interface. The purpose of these forces is to induce a certain behavior of the robot with respect to the task, such as following a reference trajectory or applying a precise interaction force. The advantage of using such guidance methods is to relieve the user’s cognitive workload during the teleoperation through proper assistance, while allowing him to control the task.
One of the problems associated with haptic guidance is to generate the task trajectories and reference forces. We need to understand what a task is and how to decompose and model it in order to generate one or more reference guides. The literature gives a few examples of task decomposition. [Wood1986] decomposes a task as the product of an action and the information derived from it. [Ramirez-Amaro2017] consider a semantic representation: each task is defined as a combination of primary actions, arbitrarily chosen.
These sub-tasks must then be associated with segmented data. This segmentation can be carried out using various methods. [Herrero2021] consider position, velocity and force data and use an Hidden Markov Model (HMM) to find the sequence of primitives (actions) associated with the subtask. [Lin2016] references and compares several of these motion segmentation methods. Once the segmentation has been performed and the primitives identified, each action must be modeled. Several models may be used based on the information that are considered, including, Dynamic Movement Primitives (DMP) ([Saveriano2021]) for motion, Contact Primitives (CP) ([Khansari2016]) and Compliant-Frame Primitives ([Galbally2022]) for interaction forces, or Probabilistic Movement Primitives (ProMP) ([Paraschos2013]) to add variability to the motion. Focusing on programming by demonstration, [Legeleux2022]’s thesis presents a comparative study of some of those modeling methods. She also presents other probabilistic methods, aimed at describing task variability, such as Gaussian Mixture Regressions (GMR) or Gaussian Mixture Models (GMM). Once this model has been obtained, it can be used to identify the operator behavior and to relate it to a reference. [Aarno2005], uses a combination of Hidden Markov Model (HMM) and Support Vector Machines (SVM) to identify the most appropriate guidance according to the user behavior. The characterization and parameterization of sub-tasks should also make it possible to add variability to task execution. Such variability could be used to modify a trajectory to avoid an obstacle, for example, or to modify the required force to perform a task in relation to a reference task.
Research activities
The internship will take place in the context of teleoperation for vertical farming. One of the tasks studied could be the opening of crop shelves, which is necessary to access the root space. We will describe this opening task and develop the needed force-motion primitives to model it. A plant photo-taking task may also be studied.
The objectives of this internship are to:
-
Identify and implement 2 to 3 modeling methods based on primitives to characterize a task. The parameterization of these models should embody and exploit variability of control primitives.
-
Apply the implemented methods to the crop shelves opening task (vertical farming shelves).
-
Identify task models from user data, acquired during previous plant photo-taking experiments which aimed at evaluate some haptic guidance.
Research environment
This research internship will take place within the Auctus team at the Inria center at the University of Bordeaux. This research team aims to meet the challenges of collaborative robotics for humans at work. The team’s research is divided into three scientific areas: analysis and modeling of human behavior (biomechanical and cognitive); human-robot interaction and coupling; design and control of cobotic systems. This internship project straddles the boundary between the second and third axes, by helping to implement and evaluate task modeling methodologies that take into account variability linked to the environment and tools, and enable robot commands to be generated. The project will draw on the hardware and software resources of the laboratory’s experimental platform. The student will be supported during his/her internship by the two supervisors.
This research internship is linked to a collaboration initiated with the IRiS lab at KAIST (Daejeon, South Korea). The student will have regular meetings with KAIST colleagues.
Skills
The candidate should have skills in : C++/Python programming, Control, and Data Processing. An additional experience in Robotics would be appreciated.
Contacts
You can send your application by email to the two supervisors with CV, cover letter and EU/Master’s grades.:
References
[Wood1986] Robert E Wood. “Task complexity: Definition of the construct”. In: Organizational Behavior and Human Decision Processes 37.1 (Feb. 1, 1986), pp. 60–82.
[Ramirez-Amaro2017] Karinne Ramirez-Amaro, Michael Beetz, and Gordon Cheng. “Transferring skills to humanoid robots by extracting semantic representations from observations of human activities”. In: Artificial Intelligence. Special Issue on AI and Robotics 247 (June 1, 2017), pp. 95–118.
[Herrero2021] Elena Galbally Herrero, Jonathan Ho, and Oussama Khatib. “Understanding and Segmenting Human Demonstrations into Reusable Compliant Primitives”. In: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Sept. 2021, pp. 9437–9444.
[Lin2016] J. F. -S. Lin, M. Karg and D. Kulić, “Movement Primitive Segmentation for Human Motion Modeling: A Framework for Analysis,” in IEEE Transactions on Human-Machine Systems, vol. 46, no. 3, pp. 325-339, June 2016.
[Saveriano2021] Matteo Saveriano et al. Dynamic Movement Primitives in Robotics: A Tutorial Survey. arXiv:2102.03861. type: article. arXiv, Feb. 7, 2021.
[Khansari2016] Mohammad Khansari, Ellen Klingbeil, et Oussama Khatib. Adaptive human-inspired compliant contact primitives to perform surface–surface contact under uncertainty. The International Journal of Robotics Research, 2016, vol. 35, no 13, p. 1651-1675.
[Galbally2022] Galbally Herrero Elena, Piedra Adrian, Brosque Cynthia, et al. Parametrization of Compliant, Object-Level Controllers from Human Demonstrations. In : International Symposium on Advances in Robot Kinematics. Cham : Springer International Publishing, 2022. p. 383-395.
[Paraschos2013] Paraschos, Alexandros et al. (2013). « Probabilistic Movement Primitives ». In : Advances in Neural Information Processing Systems. T. 26. Curran Associates, Inc.
[Legeleux2022] Amélie Legeleux. Programmation de cobots : de l’apprentissage de trajectoires à leur acceptabilité. Automatique / Robotique. Université de Bretagne Sud, 2022.
[Aarno2005] D. Aarno, S. Ekvall, and D. Kragic. “Adaptive Virtual Fixtures for Machine-Assisted Teleoperation Tasks”. In: Proceedings of the 2005 IEEE International Conference on Robotics and Automation. Apr. 2005, pp. 1139–1144.