We’re excited to report a number of new results and publications from the AUSMURI project.
Learning receptive field properties of complex cells in V1
Yanbo Lian, Tony Burkitt and collaborators. Published in PLoS Computational Biology 17(3): e1007957. https://doi.org/10.1371/journal.pcbi.1007957
This paper presents new results on learning the properties of complex cells in the primary visual cortex, as these results show that a biologically based learning model can account for the experimental data of complex cells. Together with our previous work demonstrating how simple cells in the primary visual cortex can be learnt using efficient coding, this work provides a strong basis for understanding the structure and function of the primary visual cortex, and thereby the foundation for how vision is processed in the brain.
Decentralised autonomous navigation of a UAV network for road traffic monitoring
By Andrey Savkin and Hailong Huang has been accepted by IEEE Transactions on Aerospace and Electronic Systems https://ieeexplore.ieee.org/document/9329129
With the increase of population and the fast growth of private vehicle ownership in urban areas, many roads have become more congested than ever before. The currently available traffic information collection systems mainly rely on static road-side units (RSUs), which passively record the traffic information. This paper considers an alternative option, ie, the usage of Unmanned Aerial Vehicles (UAVs) for aerial surveillance. A decentralised navigation scheme is proposed in the paper, which enables a network of UAVs to detect blockage and then gather to the blocked area to have better views of the ground vehicles. This scheme only requires the UAVs to share some light measurement information and positions, and it can be implemented in real-time.
Active Trajectory Estimation for Partially Observed Markov Decision Processes via Conditional Entropy
By Tim Molloy and Girish Nair has just been accepted for presentation and publication in the proceedings the 2021 European Control Conference (ECC), June 2021.
This paper considers the problem of controlling a partially observed Markov decision process (POMDP) in order to actively estimate its state trajectory. A novel formulation is proposed, in which the objective is to directly minimise the smoother entropy, ie, the conditional entropy of the joint state trajectory distribution of concern. This contrasts with prior approaches that minimise the sum of the conditional marginal entropies of the state estimates. By establishing a new form of the smoother entropy in terms of the POMDP belief state, the problem is reformulated as a (fully observed) Markov decision process with a value function that is concave in the belief state, making it amenable to numerical solution.
Intelligent Reference Curation for Visual Place Recognition via Bayesian Selective Fusion
Tim Molloy, Tobias Fischer, Michael Milford and Girish Nair has been accepted for publication in IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 588–595, April 2021, DOI: 10.1109/LRA.2020.3047791 and has also been accepted for parallel presentation at the 2021 IEEE Int. Conf. Robotics and Automation (ICRA), 30 May 30–5 June 2021.
We develop an algorithm that enables robots and autonomous systems to successfully calculate where they are located (a process known as localization or place recognition) using multiple references (or memories) of their environment. This work exploits shared insights from previous work conducted under the AUSMURI on place recognition using multiple temporal windows, and probabilistic approaches to resolving uncertainties in spatial awareness tasks.
Smoothing-Averse Control: Covertness and Privacy from Smoothers
T. L. Molloy and G. N. Nair, has been accepted for the 2021 American Control Conference (ACC 2021), May 2021
We investigate the problem of controlling an autonomous system such that its actions minimise the disclosure of its private information. We propose a novel information-theoretic representation of disclosed private information and develop a control-theoretic approach for sequentially minimising it. We illustrate the problem and our proposed solution in the emergent spatial awareness application of covert navigation.
Second-order Online Nonconvex Optimization
by Antoine Lesage-Landry, Joshua Taylor and Iman Shames was published in IEEE Transactions on Automatic Control on 24 November 2020. DOI: 10.1109/TAC.2020.3040372.
We present the online Newton’s method, a single-step second-order method for online nonconvex optimization. We analyze its performance and obtain a dynamic regret bound that is linear in the cumulative variation between round optima. We show that if the variation between round optima is limited, the method leads to a constant regret bound. In the general case, the online Newton's method outperforms online convex optimization algorithms for convex functions and performs similarly to a specialized algorithm for strongly convex functions. We simulate the performance of the online Newton’s method on a nonlinear, nonconvex moving target localization example and find that it outperforms a first-order approach.
Use of UAV Base Station for Searching and Bio-inspired Covert Surveillance of Tagged Wild Animals
Small tagging devices are enabling scientists to track targeted wild animals, such as migrant birds. In this paper, we introduce a prototype solution for searching and covert video surveillance of tagged wild animals based on an unmanned aerial vehicle (UAV) mounted base station (BS). We first present the concept design of a novel UAV-BS with a motion-controlled directional antenna for searching tagged wild animals in the 3-dimensional (3D) space. To reduce the disturbance of UAV to tagged wild animals, we further propose a bio-inspired motion camouflage UAV navigation algorithm for covert video surveillance of wild animals. Computer simulations show the effectiveness of the proposed method. This is the first paper on covert video surveillance of wild animals using UAVs.
Learning an efficient place cell map from grid cells using non-negative sparse coding
Yanbo Lian, Anthony N. Burkitt, “Learning an efficient place cell map from grid cells using non-negative sparse coding”, DOI: 0.1101/2020.08.12.248534
These new results on learning an efficient place map using the principle of sparse coding demonstrate the importance of sparse coding in understanding the brain: it is not only an underlying principle of processing visual information, but also for the processing of spatial information in the navigational system of the brain.
Predictive visual motion extrapolation emerges spontaneously and without supervision from a layered neural network with spike-timing-dependent plasticity
Anthony N. Burkitt, Hinze Hogendoorn, “Predictive visual motion extrapolation emerges spontaneously and without supervision from a layered neural network with spike-timing-dependent plasticity”, DOI: 10.1101/2020.08.01.232595
This paper sheds new light on our ability to track and respond to rapidly changing visual stimuli, such as a fast-moving tennis ball. This ability indicates that the brain is capable of extrapolating the trajectory of a moving object in order to predict its current position, despite the delays that result from neural transmission. In this study we show how the neural circuits underlying this ability can be learned through spike-based learning, and that the neural circuits emerge spontaneously and without supervision.
Event-based visual place recognition with ensembles of temporal windows
Tobias Fischer, Michael Milford, “Event-based visual place recognition with ensembles of temporal windows”, IEEE Robotics and Automation Letters (vol. 5, issue 4, pages 6924–6931, 2020). DOI: 10.1109/LRA.2020.3025505
In this article we dived deeper into examining the relationship between temporal intervals and place recognition performance, finding that it is highly dependent on factors like environment, conditions, camera motion and other factors. We developed a new ensemble scheme (both a full and computationally efficient ‘approximate’ version) that enabled consistently superior performance against single-window and conventional model-based ensemble approaches.
Bioinspired bearing only motion camouflage UAV guidance for covert video surveillance of a moving target
Andrey V. Savkin, Hailong Huang, “Bioinspired bearing only motion camouflage UAV guidance for covert video surveillance of a moving target”, IEEE Systems Journal, pages 1–4, 2020. [link]
This paper focuses on video monitoring of a moving ground target using an unmanned aerial vehicle (UAV). The aim is to navigate the UAV so that it is able to monitor the target while concealing its apparent motion in respect to the target’s visual system. A computationally simple sliding mode closed-loop control algorithm mimicking motion camouflage stealth behavior observed in some attacking animals is developed. The proposed guidance law is based on bearing only measurements and does not require any information on the targets’ velocity and the distance to the target. Simulations are conducted to demonstrate the effectiveness of the proposed method.