Monthly Archives: April 2018

You are browsing the site archives by month.

A nice analysis of the particularities of deep learning when applied to robotics, where the need to act is principal (unlike in other disciplines such as computer vision)

Niko Sünderhauf, Oliver Brock, Walter Scheirer, Raia Hadsell, Dieter Fox, Jürgen Leitner, Ben Upcroft, Pieter Abbeel, Wolfram Burgard, Michael Milford, and Peter Corke, The limits and potentials of deep learning for robotics, The International Journal of Robotics Research Vol 37, Issue 4-5, pp. 405 – 420, DOI: 10.1177/0278364918770733.

The application of deep learning in robotics leads to very specific problems and research questions that are typically not addressed by the computer vision and machine learning communities. In this paper we discuss a number of robotics-specific learning, reasoning, and embodiment challenges for deep learning. We explain the need for better evaluation metrics, highlight the importance and unique challenges for deep robotic learning in simulation, and explore the spectrum between purely data-driven and model-driven approaches. We hope this paper provides a motivating overview of important research directions to overcome the current limitations, and helps to fulfill the promising potentials of deep learning in robotics.

Some quotes beyond the abstract:

Deep learning systems, e.g. for classification or detection, typically return scores from their softmax layers that are proportional to the system’s confidence, but are not calibrated probabilities, and therefore not useable in a Bayesian sensor fusion framework

If, for example, an object detection system is fooled by data outside of its training data distribution (Goodfellow et al., 2014; Nguyen et al., 2015a), the consequences for a robot acting on false, but high-confidence detections can be catastrophic

As the robot moves in its environment, the camera will observe the scene from different viewpoints, which poses both challenges and opportunities to a robotic vision system […] One of the biggest advantages robotic vision can draw from its embodiment is the potential to control the camera, move it, and change its viewpoint to improve its perception or gather additional information about the scene. This is in stark contrast to most computer vision scenarios […] As an extension of active vision, a robotic system could purposefully manipulate the scene to aid its perception

In his influential 1867 book on physiological optics, Von Helmholtz (1867) formulated the idea that humans use unconscious reasoning, inference or conclusion, when processing visual information. Since then, psychologists have devised various experiments to investigate these unconscious mechanisms (Goldstein and Brockmole, 2016), modernized Helmholtz’s original ideas (Rock, 1983), and reformulated them in the framework of Bayesian inference (Kersten et al., 2004).

The properties of model-based and deep-learned approaches can be measured along multiple dimensions, including the kind of representations used for reasoning, how generally applicable their solutions are, how robust they are in real-world settings, how efficiently they make use of data, and how computationally efficient they are during operation. Model-based approaches often rely on explicit models of objects and their shape, surface, and mass properties, and use these to predict and control motion through time. In deep learning, models are typically implicitly encoded via networks and their parameters. As a consequence, model-based approaches have wide applicability, since the physics underlying them are universal. However, at the same time, the parameters of these models are difficult to estimate from perception, resulting in rather brittle performance operating only in local basins of convergence. Deep learning, on the other hand, enables highly robust performance when trained on sufficiently large data sets that are representative of the operating regime of the system. However, the implicit models learned by current deep learning techniques do not have the general applicability of physics-based reasoning. Model-based approaches are significantly more data efficient, related to their smaller number of parameters. The optimizations required for model-based approaches can be performed efficiently, but the basin of convergence can be rather small. In contrast, deep-learned solutions are often very fast and can have very large basins of convergence. However, they do not perform well if applied in a regime outside the training data

Regression to help in finding the optimal policy in MDPs based on duality theory

H. Zhu, F. Ye and E. Zhou, Solving the Dual Problems of Dynamic Programs via Regression, IEEE Transactions on Automatic Control, vol. 63, no. 5, pp. 1340-1355, DOI: 10.1109/TAC.2017.2747405.

In recent years, information relaxation and duality in dynamic programs have been studied extensively, and the resulted primal-dual approach has become a powerful procedure in solving dynamic programs by providing lower-upper bounds on the optimal value function. Theoretically, with the so-called value-based optimal dual penalty, the optimal value function could be recovered exactly via strong duality. However, in practice, obtaining tight dual bounds usually requires good approximations of the optimal dual penalty, which could be time consuming if analytical computation is not possible and nested simulation has to be used to estimate the conditional expectations inside the dual penalty. In this paper, we will develop a framework of a regression approach to approximating the optimal dual penalty in a nonnested manner, by exploring the structure of the function space consisting of all feasible dual penalties. The resulted approximations maintain to be feasible dual penalties, and thus yielding valid dual bounds on the optimal value function. We show that the proposed framework is computationally efficient, and the resulted dual penalties lead to numerically tractable dual problems. Finally, we apply the framework to a high-dimensional dynamic trading problem to demonstrate its effectiveness in solving the dual problems of complex dynamic programs.

A mathematical study of controllers that produce paths with beautfiul shapes to reach a target point by a unicycle vehicle

T. Tripathy and A. Sinha, Unicycle With Only Range Input: An Array of Patterns, IEEE Transactions on Automatic Control, vol. 63, no. 5, pp. 1300-1312, DOI: 10.1109/TAC.2017.2736940.

The objective of this paper is to generate planar patterns using an autonomous agent modeled as a unicycle. The patterns are generated about a stationary point referred to as the target. To achieve the same, the paper proposes a family of control inputs that are continuous functions of range, which is the distance between the unicycle and the target. The paper studies in detail a characterization of the resulting trajectories, which are a plethora of patterns of parametric curves (circles, spirals, epicyclic curves like hypotrochoids) and more. These appealing patterns find applications in exploration, coverage, land mine detection, etc., where the target represents any point of interest like a landmark or a beacon. The paper also investigates the necessary conditions on the control laws in order to generate patterns of desired shapes and bounds. Furthermore, to generate desired patterns with arbitrary initial conditions, a switching strategy is proposed which is illustrated using an algorithm. The paper presents a series of simulations of appealing patterns generated using the proposed control laws.

How a robot can learn to recognize itself on a mirror

Zeng, Y., Zhao, Y., Bai, J. et al., Toward Robot Self-Consciousness (II): Brain-Inspired Robot Bodily Self Model for Self-Recognition, Cogn Comput (2018) 10: 307, DOI: 10.1007/s12559-017-9505-1.

The neural correlates and nature of self-consciousness is an advanced topic in Cognitive Neuroscience. Only a few animal species have been testified to be with this cognitive ability. From artificial intelligence and robotics point of view, few efforts are deeply rooted in the neural correlates and brain mechanisms of biological self-consciousness. Despite the fact that the scientific understanding of biological self-consciousness is still in preliminary stage, we make our efforts to integrate and adopt known biological findings of self-consciousness to build a brain-inspired model for robot self-consciousness. In this paper, we propose a brain-inspired robot bodily self model based on extensions to primate mirror neuron system and apply it to humanoid robot for self recognition. In this model, the robot firstly learns the correlations between self-generated actions and visual feedbacks in motion by learning with spike timing dependent plasticity (STDP), and then learns the appearance of body part with the expectation that the visual feedback is consistent with its motion. Based on this model, the robot uses multisensory integration to learn its own body in real world and in mirror. Then it can distinguish itself from others. In a mirror test setting with three robots with the same appearance, with the proposed brain-inspired robot bodily self model, each of them can recognize itself in the mirror after these robots make random movements at the same time. The theoretic modeling and experimental validations indicate that the brain-inspired robot bodily self model is biologically inspired, and computationally feasible as a foundation for robot self recognition.

Methods for estimating periods of noisy signals

W. Fan, Y. Li, K. L. Tsui and Q. Zhou, A Noise Resistant Correlation Method for Period Detection of Noisy Signals, IEEE Transactions on Signal Processing, vol. 66, no. 10, pp. 2700-2710, DOI: 10.1109/TSP.2018.2813305.

This paper develops a novel method called the noise resistant correlation method for detecting the hidden period from the contaminated (noisy) signals with strong white Gaussian noise. A novel correlation function is proposed based on a newly constructed periodic signal and the contaminated signal to effectively detect the target hidden period. In contrast with the conventional autocorrelation analysis (AUTOC) method, this method demonstrates excellent performance, especially when facing strong noise. Fault diagnoses of rolling element bearings and gears are presented as application examples and the performance of the proposed method is compared with that of the AUTOC method.

Another paper about this in the same issue: 10.1109/TSP.2018.2818080.

A novel algorithm for coverage path planning with very strong guarantees

J. Song and S. Gupta, $varepsilon ^{star }$: An Online Coverage Path Planning Algorithm, IEEE Transactions on Robotics, vol. 34, no. 2, pp. 526-533, DOI: 10.1109/TRO.2017.2780259.

This paper presents an algorithm called ε*, for online coverage path planning of unknown environment. The algorithm is built upon the concept of an Exploratory Turing Machine (ETM), which acts as a supervisor to the autonomous vehicle to guide it with adaptive navigation commands. The ETM generates a coverage path online using Multiscale Adaptive Potential Surfaces (MAPS), which are hierarchically structured and dynamically updated based on sensor information. The ε*-algorithm is computationally efficient, guarantees complete coverage, and does not suffer from the local extrema problem. Its performance is validated by 1) high-fidelity simulations on Player/Stage and 2) actual experiments in a laboratory setting on autonomous vehicles.

Rao-Blackwellized Particle Filter SLAM with grid maps in which particles do not contain the whole map but only a part

H. Jo, H. M. Cho, S. Jo and E. Kim, Efficient Grid-Based Rao–Blackwellized Particle Filter SLAM With Interparticle Map Sharing, IEEE/ASME Transactions on Mechatronics, vol. 23, no. 2, pp. 714-724, DOI: 10.1109/TMECH.2018.2795252.

In this paper, we propose a novel and efficient grid-based Rao-Blackwellized particle filter simultaneous localization and mapping (RBPF-SLAM) with interparticle map shaping (IPMS). The proposed method aims at saving the computational memory in the grid-based RBPF-SLAM while maintaining the mapping accuracy. Unlike conventional RBPF-SLAM in which each particle has its own map of the whole environment, each particle has only a small map of the nearby environment called an individual map in the proposed method. Instead, the map of the remaining large environment is shared by the particles. The part shared by the particles is called a base map. If the individual small maps become reliable enough to trust, they are merged with the base map. To determine when and which part of an individual map should be merged with the base map, we propose two map sharing criteria. Finally, the proposed IPMS RBPF-SLAM is applied to the real-world datasets and benchmark datasets. The experimental results show that our method outperforms conventional methods in terms of map accuracy versus memory consumption.

What is Cognitive Computational Neuroscience

Thomas Naselaris, Danielle S. Bassett, Alyson K. Fletcher, Konrad Kording, Nikolaus Kriegeskorte, Hendrikje Nienborg, Russell A. Poldrack, Daphna Shohamy, Kendrick Kay, Cognitive Computational Neuroscience: A New Conference for an Emerging Discipline, Trends in Cognitive Sciences, Volume 22, Issue 5, 2018, Pages 365-367, DOI: 10.1016/j.tics.2018.02.008.

Understanding the computational principles that underlie complex behavior is a central goal in cognitive science, artificial intelligence, and neuroscience. In an attempt to unify these disconnected communities, we created a new conference called Cognitive Computational Neuroscience (CCN). The inaugural meeting revealed considerable enthusiasm but significant obstacles remain.

Robot topological navigation

Sergio Miguel-Tomé, Navigation through unknown and dynamic open spaces using topological notions,Connection Science vol. 30, iss. 2, DOI: 10.1080/09540091.2016.1277691.

Until now, most algorithms used for navigation have had the purpose of directing system towards one point in space. However, humans communicate tasks by specifying spatial relations among elements or places. In addition, the environments in which humans develop their activities are extremely dynamic. The only option that allows for successful navigation in dynamic and unknown environments is making real-time decisions. Therefore, robots capable of collaborating closely with human beings must be able to make decisions based on the local information registered by the sensors and interpret and express spatial relations. Furthermore, when one person is asked to perform a task in an environment, this task is communicated given a category of goals so the person does not need to be supervised. Thus, two problems appear when one wants to create multifunctional robots: how to navigate in dynamic and unknown environments using spatial relations and how to accomplish this without supervision. In this article, a new architecture to address the two cited problems is presented, called the topological qualitative navigation architecture. In previous works, a qualitative heuristic called the heuristic of topological qualitative semantics (HTQS) has been developed to establish and identify spatial relations. However, that heuristic only allows for establishing one spatial relation with a specific object. In contrast, navigation requires a temporal sequence of goals with different objects. The new architecture attains continuous generation of goals and resolves them using HTQS. Thus, the new architecture achieves autonomous navigation in dynamic or unknown open environments.

A novel fast algorithm for clock synchronization in a wireless network, with a nice introduction but assuming negligible communication times and thus not directly applicable in teleoperation

Kan Xie, Qianqian Cai, Minyue Fu, A fast clock synchronization algorithm for wireless sensor networks, Automatica, Volume 92, 2018, Pages 133-142, DOI: 10.1016/j.automatica.2018.03.004.

This paper proposes a novel clock synchronization algorithm for wireless sensor networks (WSNs). The algorithm is derived using a fast finite-time average consensus idea, and is fully distributed, meaning that each node relies only on its local clock readings and reading announcements from its neighbours. For networks with an acyclic graph, the algorithm converges in only d iterations for clock rate synchronization and another d iterations for clock offset synchronization, where d is the graph diameter. The algorithm enjoys low computational and communicational complexities and robustness against transmission adversaries. Each node can execute the algorithm asynchronously without the need for global coordination. Due to its fast convergence, the algorithm is most suitable for large-scale WSNs. For WSNs with a cyclic graph, a fast distributed depth-first-search (DFS) algorithm can be applied first to form a spanning tree before applying the proposed synchronization algorithm.