Multi-contact tactile exploration and interaction with unknown objects

Humans rely on the sense of touch in almost every aspect of daily life, whether to tie shoelaces, place fingertips on a computer keyboard or find keys inside a bag. With robots moving into human-centered environment, tactile exploration becomes more and more important as vision may be occluded easily by obstacles or fail because of different illumination conditions. Traditional approaches mostly rely on position control for manipulating objects and are adapted to single grippers and known objects. New sensors make it possible to extend the control to tackle problems unsolved before: handling unknown objects and discovering local features on their surface. This thesis tackles the problem of controlling a robot which makes multiple contacts with an unknown environment. Generating and keeping multiple contacts points on different parts of the robot fingers during exploration is an essential feature that distinguishes our work from other haptic exploration work in the literature, where contacts are usually limited to one or more fingertips. In the first part of this thesis, we address the problem of exploring partially known surfaces and objects for modeling and identification. In multiple scenarios, control and exploration strategies are developed to compliantly follow the surface or contour of a surface with robotic fingers. Whereas the methods developed in the first part of this thesis perform well on objects with limited size and variation in shape, the second part of the thesis is devoted to the development of a controller that maximizes contact with unknown surfaces of any shape and size. Maximizing contact allows to gather information more rapidly and also to create stable grasps. To this end, we develop an algorithm based on the task-space formulation to quickly handle the control in torque of an actively compliant robot while keeping constraints, particularly on contact forces. We also develop a strategy to maximize the surface in contact, given only the current state of contact, i.e. without prior information on the object or surface. In the third part of the thesis, an additional application of the developed hand controller is explored. The problem of autonomous grasping using only tactile data is tackled. The arm motion is generated according to search and grasping strategies implemented with Dynamical Systems (DS). We extend existing approaches to locally modulate dynamical systems (DS) to enable sensing-based modulation, so as to change the dynamics of motion depending on task progress. This allows to generate fast and autonomous object localization and grasping in one flexible framework. We also apply this algorithm to teach a robot how to react to collisions in order to navigate between obstacles while reaching.

Grasping for the Task

Ravin Luis De Souza

The significant advances made in the design and construction of anthropomorphic robot hands, endow them with prehensile abilities reaching that of humans. However, using these powerful hands with the same level of expertise that humans display is a big challenge for robots. Traditional approaches use finger-tip (precision) or enveloping (power) methods to generate the best force closure grasps. However, this ignores the variety of prehensile postures available to the hand and also the larger context of arm action. This thesis explores a paradigm for grasp formation based on generating oppositional pressure within the hand, which has been proposed as a functional basis for grasping in humans (MacKenzie and Iberall, 1994). A set of opposition primitives encapsulates the hand's ability to generate oppositional forces. The oppositional intention encoded in a primitive serves as a guide to match the hand to the object, quantify its functional ability and relate this to the arm. In this thesis we leverage the properties of opposition primitives to both interpret grasps formed by humans and to construct grasps for a robot considering the larger context of arm action. In the first part of the thesis we examine the hypothesis that hand representation schemes based on opposition are correlated with hand function. We propose hand-parameters describing oppositional intention and compare these with commonly used methods such as joint angles, joint synergies and shape features. We expect that opposition-based parameterizations, which take an interaction-based perspective of a grasp, are able to discriminate between grasps that are similar in shape but different in functional intent. We test this hypothesis using qualitative assessment of precision and power capabilities found in existing grasp taxonomies. The next part of the thesis presents a general method to recognize oppositional intention manifested in human grasp demonstrations. A data glove instrumented with tactile sensors is used to provide the raw information regarding hand configuration and interaction force. For a grasp combining several cooperating oppositional intentions, hand surfaces can be simultaneously involved in multiple oppositional roles. We characterize the low-level interactions between different surfaces of the hand based on captured interaction force and reconstructed hand surface geometry. This is subsequently used to separate out and prioritize multiple and possibly overlapping oppositional intentions present in the demonstrated grasp. We evaluate our method on several human subjects across a wide range of hand functions. The last part of the thesis applies the properties encoded in opposition primitives to optimize task performance of the arm, for tasks where the arm assumes the dominant role. For these tasks, choosing the strongest power grasp available (from a force-closure sense) may constrain the arm to a sub-optimal configuration. Weaker grasp components impose fewer constraints on the hand, and can therefore explore a wider region of the object relative pose space. We take advantage of this to find the good arm configurations from a task perspective. The final hand-arm configuration is obtained by trading of overall robustness in the grasp with ability of the arm to perform the task. We validate our approach, using the tasks of cutting, hammering, screw-driving and opening a bottle-cap, for both human and robotic hand-arm systems.

EPFL2016

Imitation Learning of Motion Coordination in Robots

Elena Gribovskaya

The ease with which humans coordinate all their limbs is fascinating. Such a simplicity is the result of a complex process of motor coordination, i.e. the ability to resolve the biomechanical redundancy in an efficient and repeatable manner. Coordination enables a wide variety of everyday human activities from filling in a glass with water to pair figure skating. Therefore, it is highly desirable to endow robots with similar skills. Despite the apparent diversity of coordinated motions, all of them share a crucial similarity: these motions are dictated by underlying constraints. The constraints shape the formation of the coordination patterns between the different degrees of freedom. Coordination constraints may take a spatio-temporal form; for instance, during bimanual object reaching or while catching a ball on the fly. They also may relate to the dynamics of the task; for instance, when one applies a specific force profile to carry a load. In this thesis, we develop a framework for teaching coordination skills to robots. Coordination may take different forms, here, we focus on teaching a robot intra-limb and bimanual coordination, as well as coordination with a human during physical collaborative tasks. We use tools from well-established domains of Bayesian semiparametric learning (Gaussian Mixture Models and Regression, Hidden Markov Models), nonlinear dynamics, and adaptive control. We take a biologically inspired approach to robot control. Specifically, we adopt an imitation learning perspective to skill transfer, that offers a seamless and intuitive way of capturing the constraints contained in natural human movements. As the robot is taught from motion data provided by a human teacher, we exploit evidence from human motor control of the temporal evolution of human motions that may be described by dynamical systems. Throughout this thesis, we demonstrate that the dynamical system view on movement formation facilitates coordination control in robots. We explain how our framework for teaching coordination to a robot is built up, starting from intra-limb coordination and control, moving to bimanual coordination, and finally to physical interaction with a human. The dissertation opens with the discussion of learning discrete task-level coordination patterns, such as spatio-temporal constraints emerging between the two arms in bimanual manipulation tasks. The encoding of bimanual constraints occurs at the task level and proceeds through a discretization of the task as sequences of bimanual constraints. Once the constraints are learned, the robot utilizes them to couple the two dynamical systems that generate kinematic trajectories for the hands. Explicit coupling of the dynamical systems ensures accurate reproduction of the learned constraints, and proves to be crucial for successful accomplishment of the task. In the second part of this thesis, we consider learning one-arm control policies. We present an approach to extracting non-linear autonomous dynamical systems from kinematic data of arbitrary point-to-point motions. The proposed method aims to tackle the fundamental questions of learning robot coordination: (i) how to infer a motion representation that captures a multivariate coordination pattern between degrees of freedom and that generalizes this pattern to unseen contexts; (ii) whether the policy learned directly from demonstrations can provide robustness against spatial and temporal perturbations. Finally, we demonstrate that the developed dynamical system approach to coordination may go beyond kinematic motion learning. We consider physical interactions between a robot and a human in situations where they jointly perform manipulation tasks; in particular, the problem of collaborative carrying and positioning of a load. We extend the approach proposed in the second part of this thesis to incorporate haptic information into the learning process. As a result, the robot adapts its kinematic motion plan according to human intentions expressed through the haptic signals. Even after the robot has learned the task model, the human still remains a complex contact environment. To ensure robustness of the robot behavior in the face of the variability inherent to human movements, we wrap the learned task model in an adaptive impedance controller with automatic gain tuning. The techniques, developed in this thesis, have been applied to enable learning of unimanual and bimanual manipulation tasks on the robotics platforms HOAP-3, KATANA, and i-Cub, as well as to endow a pair of simulated robots with the ability to perform a manipulation task in the physical collaboration.

EPFL2012

Grasping for the Task

Ravin Luis De Souza

EPFL2016

Imitation Learning of Motion Coordination in Robots

Elena Gribovskaya

EPFL2012