Online spatio-temporal Gaussian process experts with application to tactile classification

Abstract

Abstract-In this work, we are primarily concerned with robotic systems that learn online and continuously from multivariate data-streams. Our first contribution is a new recursive kernel, which we have integrated into a sparse Gaussian Process to yield the Spatio-Temporal Online Recursive Kernel Gaussian Process (STORK-GP). This algorithm iteratively learns from time-series, providing both predictions and uncertainty estimates. Experiments on benchmarks demonstrate that our method achieves high accuracies relative to state-of-the-art methods. Second, we contribute an online tactile classifier which uses an array of STORK-GP experts. In contrast to existing work, our classifier is capable of learning new objects as they are presented, improving itself over time. We show that our approach yields results comparable to highly-optimised offline classification methods. Moreover, we conducted experiments with human subjects in a similar online setting with true-label feedback and present the insights gained.

When and how to help: An iterative probabilistic model for learning assistance by demonstration

“…Since standard GPs are computationally expensive (requiring O(n 3 ) time for training and O(n 2 ) space) and we typically prefer our robotic assistants to learn iteratively, we used the Spatio-Temporal Online Recursive Kernel Gaussian Process (STORK-GP) [12], a specialised sparse GP with memory to account for temporal dependencies. In the following subsections, we flesh out our model by giving specifics on how our model can be trained efficiently in real-time.…”

“…In contrast to prevailing methods, the STORK-GP can be updated sequentially in real-time and models temporal dependencies. Because its internals are relatively involved and discussed elsewhere [17], [12], we focus on describing the three main aspects of the algorithm and provide references for readers wanting additional detail:…”

“…where M is a symmetric d ⇥ d matrix that controls the "importance" of the inputs [12]. It is conventional that…”

Spatio-Temporal Learning With the Online Finite and Infinite Echo-State Gaussian Processes

“…Both methods operate on noisy multivariate time-series data on a fixed maximum computational and storage budget to produce high accuracies relative to state-of-the-art methods. This paper collates and extends [1] and [2] with iterative hyperparameter adaptation, in-depth derivation and analysis, and new experiments. Furthermore, we present two case studies in online robot learning-by-demonstration (LbD) involving the Nao humanoid robot and the ARTY smart wheelchair.…”

Online quantum mixture regression for trajectory learning by demonstration

“…For the locally weighted projection regression [12], we have used the authors' LWPR library [28], [29]. For the online Gaussian Process (oGP), we utilize a matlab version of the OTL library that appears in [30], [31]. Testing was conducted in Matlab 2012b on an Ubuntu Linux PC, i7 3.4GHz, 16GB RAM.…”