Iterative learning of grasp adaptation through human corrections

Abstract

Abstract-In the context of object interaction and manipulation, one characteristic of a robust grasp is its ability to comply with external perturbations applied to the grasped object while still maintaining the grasp. In this work we introduce an approach for grasp adaptation which learns a statistical model to adapt hand posture solely based on the perceived contact between the object and fingers. Using a multi-step learning procedure, the model dataset is built by first demonstrating an initial hand posture, which is then physically corrected by a human teacher pressing on the fingertips, exploiting compliance in the robot hand. The learner then replays the resulting sequence of hand postures, to generate a dataset of posture-contact pairs that are not influenced by the touch of the teacher. A key feature of this work is that the learned model may be further refined by repeating the correction-replay steps. Alternatively, the model may be reused in the development of new models, characterized by the contact signatures of a different object. Our approach is empirically validated on the iCub robot. We demonstrate grasp adaptation in response to changes in contact, and show successful model reuse and improved adaptation with additional rounds of model refinement.

Probabilistic consolidation of grasp experience

“…Compared to our work, this system did not integrate experience from training data or feedback from grasp execution. Differently from [10], [11] and [8] included an off-line training phase based on examples demonstrated by a teacher. In our work, training also relies on human demonstration.…”

“…Different approaches have been studied, e.g., analytic [2] and data-driven [4]. Moreover, different subproblems have been addressed, e.g., grasp planning [5], force control [6], stability estimation from sensory data after grasp execution [7] or grasp adaptation [8]. However, current robotic systems still have severe limitations in dealing with novelty, uncertainty and unforeseen situations.…”

Learning a real time grasping strategy

“…A Gaussian Mixture Model (GMM) is used here to get a probabilistic encoding of the joint distribution p(h,o,θ | Ω ). We choose to use GMM because of its ability to effectively extrapolate the missing data, as has been exploited in many applications [1], [23]. It also has the advantage of capturing the non-linearity of the space, as well as determining how likely a point in the input space is under the model.…”

“…Secondly,N N N is a normalized function bounded between 0 and 1. This ensures the points with the same Mahalanobis distance from a Gaussian will have the same membership value, regardless of the size of the covariance [23].…”

Continuing Robot Skill Learning after Demonstration with Human Feedback

“…Tactile feedback is used to assist in both policy refinement and the reuse of a demonstrated policy when developing a different policy; effectively using the demonstrated policy as prior knowledge for a new behavior. Empirical validation has included grasp positioning on the iCub humanoid [4], as well as grasp adaptation in response to changes in fingertip contact [13].…”