This paper proposes an end-to-end learning from demonstration framework for teaching force-based manipulation tasks to robots. The strengths of this work are manyfold. First, we deal with the problem of learning through force perceptions exclusively. Second, we propose to exploit haptic feedback both as a means for improving teacher demonstrations and as a human–robot interaction tool, establishing a bidirectional communication channel between the teacher and the robot, in contrast to the works using kinesthetic teaching. Third, we address the well-known what to imitate? problem from a different point of view, based on the mutual information between perceptions and actions. Lastly, the teacher’s demonstrations are encoded using a Hidden Markov Model, and the robot execution phase is developed by implementing a modified version of Gaussian Mixture Regression that uses implicit temporal information from the probabilistic model, needed when tackling tasks with ambiguous perceptions. Experimental results show that the robot is able to learn and reproduce two different manipulation tasks, with a performance comparable to the teacher’s one.
Robot learning from demonstration faces new challenges when applied to tasks in which forces play a key role. Pouring liquid from a bottle into a glass is one such task, where not just a motion with a certain force profile needs to be learned, but the motion is subtly conditioned by the amount of liquid in the bottle. In this paper, the pouring skill is taught to a robot as follows. In a training phase, the human teleoperates the robot using a haptic device, and data from the demonstrations are statistically encoded by a parametric hidden Markov model, which compactly encapsulates the relation between the task parameter (dependent on the bottle weight) and the force-torque traces. Gaussian mixture regression is then used at the reproduction stage for retrieving the suitable robot actions based on the force perceptions. Computational and experimental results show that the robot is able to learn to pour drinks using the proposed framework, outperforming other approaches such as the classical hidden Markov models in that it requires less training, yields more compact encodings and shows better generalization capabilities.
Research in learning from demonstration has focused on transferring movements from humans to robots. However, a need is arising for robots that do not just replicate the task on their own, but that also interact with humans in a safe and natural way to accomplish tasks cooperatively. Robots with variable impedance capabilities opens the door to new challenging applications, where the learning algorithms must be extended by encapsulating force and vision information. In this paper we propose a framework to transfer impedance-based behaviors to a torque-controlled robot by kinesthetic teaching. The proposed model encodes the exam- ples as a task-parameterized statistical dynamical system, where the robot impedance is shaped by estimating virtual stiffness matrices from the set of demonstrations. A collaborative assembly task is used as testbed. The results show that the model can be used to modify the robot impedance along task execution to facilitate the collaboration, by triggering stiff and compliant behaviors in an on-line manner to adapt to the user's actions.
The GARNICS project aims at 3D sensing of plant growth and building perceptual representations for learning the links to actions of a robot gardener. Plants are complex, self-changing systems with increasing complexity over time. Actions performed at plants (like watering), will have strongly delayed effects. Thus, monitoring and controlling plants is a difficult perception-action problem requiring advanced predictive cognitive properties, which so far can only be provided by experienced human gardeners. Sensing and control of a plants actual properties, i.e. its phenotype, is relevant to e.g. seed production and plant breeders. We address plant sensing and control by combining active vision with appropriate perceptual representations, which are essential for cognitive interactions. Core ingredients for these representations are channel representations, dynamic graphs and cause-effect couples (CECs). Channel representations are a wavelet-like, biologically motivated information representation, which can be generalized coherently using group theory. Using these representations, plant models -- represented by dynamic graphs -- will be acquired and by interacting with a human gardener the system will be taught the different cause-effect relations resulting from possible treatments. Employing decision making and planning processes via CECs, our robot gardener will then be able to choose from its learned repertoire the appropriate actions for optimal plant growth. This way we will arrive at an adaptive, interactive cognitive system, which will be implemented and tested in an industrially-relevant plant-phenotyping application.
Rozo, L.; Jimenez, P.; Torras, C. International Conference of the Catalan Association for Artificial Intelligence p. 331-341 DOI: 10.3233/978-1-60750-643-0-331 Presentation's date: 2010 Presentation of work at congresses
Locally weighted learning algorithms are suitable strategies for trajectory learning and skill acquisition, in the context of programming by demonstration. Input streams other than visual information, as used in most applications up to date, reveal themselves as quite useful in trajectory learning experiments where visual sources are not available. In this work we have
used force/torque feedback through a haptic device for teaching a teleoperated robot to empty a rigid container. Structure vibrations and container inertia appeared to considerably disrupt the sensing process, so a filtering algorithm had to be devised. Then, the memory-based LWPLS and the non-memory-based LWPR algorithms [8, 13, 10] were implemented, their comparison
leading to very similar results, with the same pattern as regards to both the involved robot joints and the different initial experimental conditions. Tests where the teacher was instructed to follow a strategy compared to others where he was not lead to useful conclusions that permit devising the new research stages, where the taught motion will be refined by autonomous robot rehearsal through reinforcement learning.