Directing robot attention to recognise activities and to anticipate events like goal-directed actions is a crucial skill for human-robot interaction. Unfortunately, issues like intrinsic time constraints, the spatially distributed nature of the entailed information sources, and the existence of a multitude of unobservable states affecting the system, like latent intentions, have long rendered achievement of such skills a rather elusive goal. The problem tests the limits of current attention control systems. It requires an integrated solution for tracking, exploration and recognition, which traditionally have been seen as separate problems in active vision.We propose a probabilistic generative framework based on a mixture of Kalman filters and information gain maximisation that uses predictions in both recognition and attention-control. This framework can efficiently use the observations of one element in a dynamic environment to provide information on other elements, and consequently enables guided exploration.Interestingly, the sensors-control policy, directly derived from first principles, represents the intuitive trade-off between finding the most discriminative clues and maintaining overall awareness.Experiments on a simulated humanoid robot observing a human executing goal-oriented actions demonstrated improvement on recognition time and precision over baseline systems.

We introduce a novel algorithm called Upper Confidence Weighted Learning (UCWL) for online multiclass learning from binary feedback. UCWL combines the Upper Confidence Bound (UCB) framework with the Soft Confidence Weighted (SCW) online learning scheme.  UCWL achieves state of the art performance (especially on noisy and non-separable data) with low computational costs. Estimated confidence intervals are used for informed exploration, which enables faster learning than the uninformed exploration case or the case where exploration is not used.  The targeted application setting is human-robot interaction (HRI), in which a robot is learning to classify its observations while a human teaches it by providing only binary feedback (e.g., right/wrong).  Results in an HRI experiment, and with two benchmark datasets, show UCWL outperforms other algorithms in the online binary feedback setting, and surprisingly even sometimes beats state-of-the-art algorithms that get full feedback, while UCWL gets only binary feedback on the same data.

Localization of a mobile robot is crucial for autonomous navigation. Using laser scanners, this can be facilitated by the pairwise alignment of consecutive scans. In this paper, we are interested in improving this scan alignment in challenging natural environments. For this purpose, local descriptors are generally effective as they facilitate point matching. However, we show that in some natural environments, many of them are likely to be unreliable, which affects the accuracy and robustness of the results. Therefore, we propose to filter the unreliable descriptors as a prior step to alignment. Our approach uses a fast machine learning algorithm, trained on-the-fly under the positive and unlabeled learning paradigm without the need for human intervention. Our results show that the number of descriptors can be significantly reduced, while increasing the proportion of reliable ones, thus speeding up and improving the robustness of the scan alignment process.

Dispersing a team of robots into an unknown and dangerous environment, such as a collapsed building, can provide information about structural damage and locations of survivors and help rescuers plan their actions. We propose a rolling dispersion algorithm, which makes use of a small number of robots and achieves full exploration. The robots disperse as much as possible while maintaining communication, and then advance as a group, leaving behind beacons to mark explored areas and provide a path back to the entrance. The novelty of this algorithm comes from the manner in which the robots continue their exploration as a group after reaching the maximum dispersion possible while staying in contact with each other. We use simulation to show that the algorithm works in multiple environments and for varying numbers of robots.

Human action recognition from videos is a challenging machine vision task with multiple important application domains, such as human-robot/machine interaction, interactive entertainment, multimedia information retrieval, and surveillance. In this paper, we present a novel approach to human action recognition from 3D skeleton sequences extracted from depth data. We use the covariance matrix for skeleton joint locations over time as a discriminative descriptor for a sequence. To encode the relationship between joint movement and time, we deploy multiple covariance matrices over sub-sequences in a hierarchical fashion. The descriptor has a fixed length that is independent from the length of the described sequence. Our experiments show that using the covariance descriptor with an off-the-shelf classification algorithm outperforms the state of the art in action recognition on multiple datasets, captured either via a Kinect-type sensor or a sophisticated motion capture system. We also include an evaluation on a novel large dataset using our own annotation.

Users may ask a service robot to accomplish various tasks so that the designer of the robot cannot program each of the tasks beforehand. As more and more open-source knowledge resources become available, it is worthwhile trying to make use of open-source knowledge resources for service robots. The challenge lies in the autonomous identification, acquisition and utilization of missing knowledge about a user task at hand. In this paper, the core problem is formalized and the complexity results of the main reasoning issues are provided. A mechanism for task planning with open-knowledge rules which are provided by non-experts in semi-structured natural language and thus generally underspecified are introduced. Techniques for translating the semi-structured knowledge from a large open-source knowledge base are also presented. Experiments showed a remarkable improvement of the system performance on a test set consisting of hundreds of user desires from the open-source knowledge base.

Most existing work on learning planning models assumes that the entire model needs to be learned from scratch. A more realistic situation is that the planning agent has an incomplete model which it needs to refine through learning. In this paper we propose and evaluate a method for doing this. Our method takes as input an incomplete model (with missing preconditions and effects in the actions), as well as a set of plan traces that are known to be correct. It outputs a refined model that not only captures additional precondition/effect knowledge about the given actions, but also macro actions. We use a MAX-SAT framework for learning, where the constraints are derived from the executability of the given plan traces, as well as the preconditions/effects of the given incomplete model. Unlike traditional macro-action learners which use macros to increase the efficiency of planning (in the context of a complete model), our motivation for learning macros is to increase the accuracy (robustness) of the plans generated with the refined model. We demonstrate the effectiveness of our approach through a systematic empirical evaluation.

There is increasing awareness in the planning community that the burden of specifying complete domain models is too high, which impedes the applicability of planning technology in many real-world domains. Although there have been many learning approaches that help automatically creating domain models, they all assume plan traces (training data) are \emph{correct}. In this paper, we aim to remove this assumption, allowing plan traces to be with noise. Compared to collecting large amount of correct plan traces, it is much easier to collect noisy plan traces, e.g., we can directly exploit sensors to help collect noisy plan traces. We consider a novel solution for this challenge that can learn action models from noisy plan traces. We create a set of random variables to capture the possible correct plan traces behind the observed noisy ones, and build a graphical model to describe the physics of the domain. We then learn the parameters of the graphical model and acquire the domain model based on the learnt parameters. In the experiment, we empirically show that our approach is effective in several planning domains.

Recent advances in solutions to Hybrid MDPs with discrete and continuous state and action spaces have significantly extended the class of MDPs for which exact solutions can be derived, albeit at the expense of a restricted transition noise model. In this paper, we work around limitations of previous solutions by adopting a robust optimization approach in which Nature is allowed to adversarially determine transition noise within pre-specified confidence intervals. This allows one to derive an optimal policy with an arbitrary (user-specified) level of success probability and significantly extends the class of transition noise models for which Hybrid MDPs can be solved. This work also significantly extends results for the related ``chance-constrained'' approach in stochastic hybrid control to accommodate state-dependent noise. We demonstrate our approach working on a variety of hybrid MDPs taken from AI planning, operations research, and control theory, noting that this is the first time robust solutions with strong guarantees over all states have been automatically derived for such problems.

Over-constrained temporal problems are commonly encountered while operating autonomous and decision support systems. An intelligent system must learn a human's preference over a problem in order to generate preferred resolutions that minimize perturbation. We present the Best-first Conflict-Directed Relaxation (BCDR) algorithm for enumerating the best continuous relaxation for an over-constrained conditional temporal problem with controllable choices. BCDR reformulates such a problem by making its temporal constraints relaxable and solves the problem using a conflict-directed approach. It extends the Conflict-Directed A* (CD-A*) algorithm to conditional temporal problems, by first generalizing the conflict learning process to include all discrete variable assignments and continuous temporal constraints, and then by guiding the forward search away from known infeasible regions using conflict resolution. When evaluated empirically on a range of coordinated car sharing network problems, BCDR demonstrates a substantial improvement in performance and solution quality compared to previous conflict-directed approaches.