Monthly Archives: June 2019

You are browsing the site archives by month.

Enforcing safe behaviour on critical systems that use machine learning through robust control and bayesian inference

J. F. Fisac, A. K. Akametalu, M. N. Zeilinger, S. Kaynama, J. Gillula and C. J. Tomlin, A General Safety Framework for Learning-Based Control in Uncertain Robotic Systems, IEEE Transactions on Automatic Control, vol. 64, no. 7, pp. 2737-2752 DOI: 10.1109/TAC.2018.2876389.

The proven efficacy of learning-based control schemes strongly motivates their application to robotic systems operating in the physical world. However, guaranteeing correct operation during the learning process is currently an unresolved issue, which is of vital importance in safety-critical systems. We propose a general safety framework based on Hamilton–Jacobi reachability methods that can work in conjunction with an arbitrary learning algorithm. The method exploits approximate knowledge of the system dynamics to guarantee constraint satisfaction while minimally interfering with the learning process. We further introduce a Bayesian mechanism that refines the safety analysis as the system acquires new evidence, reducing initial conservativeness when appropriate while strengthening guarantees through real-time validation. The result is a least-restrictive, safety-preserving control law that intervenes only when the computed safety guarantees require it, or confidence in the computed guarantees decays in light of new observations. We prove theoretical safety guarantees combining probabilistic and worst-case analysis and demonstrate the proposed framework experimentally on a quadrotor vehicle. Even though safety analysis is based on a simple point-mass model, the quadrotor successfully arrives at a suitable controller by policy-gradient reinforcement learning without ever crashing, and safely retracts away from a strong external disturbance introduced during flight.

Use of Markov Decision Processes to select tasks in a service mobile robots

Lacerda, B., Faruq, F., Parker, D., & Hawes, N., Probabilistic planning with formal performance guarantees for mobile service robots, The International Journal of Robotics Research, DOI: 10.1177/0278364919856695.

We present a framework for mobile service robot task planning and execution, based on the use of probabilistic verification techniques for the generation of optimal policies with attached formal performance guarantees. Our approach is based on a Markov decision process model of the robot in its environment, encompassing a topological map where nodes represent relevant locations in the environment, and a range of tasks that can be executed in different locations. The navigation in the topological map is modeled stochastically for a specific time of day. This is done by using spatio-temporal models that provide, for a given time of day, the probability of successfully navigating between two topological nodes, and the expected time to do so. We then present a methodology to generate cost optimal policies for tasks specified in co-safe linear temporal logic. Our key contribution is to address scenarios in which the task may not be achievable with probability one. We introduce a task progression function and present an approach to generate policies that are formally guaranteed to, in decreasing order of priority: maximize the probability of finishing the task; maximize progress towards completion, if this is not possible; and minimize the expected time or cost required. We illustrate and evaluate our approach with a scalability evaluation in a simulated scenario, and report on its implementation in a robot performing service tasks in an office environment for long periods of time.

On theories of human decision making and the role of affects

Ian D. Roberts, Cendri A. Hutcherson, Affect and Decision Making: Insights and Predictions from Computational Models, Trends in Cognitive Sciences,
Volume 23, Issue 7, 2019, Pages 602-614 DOI: 10.1016/j.tics.2019.04.005.

In recent years interest in integrating the affective and decision sciences has skyrocketed. Immense progress has been made, but the complexities of each field, which can multiply when combined, present a significant obstacle. A carefully defined framework for integration is needed. The shift towards computational modeling in decision science provides a powerful basis and a path forward, but one whose synergistic potential will only be fully realized by drawing on the theoretical richness of the affective sciences. Reviewing research using a popular computational model of choice (the drift diffusion model), we discuss how mapping concepts to parameters reduces conceptual ambiguity and reveals novel hypotheses.

On the not clear distinction between fast/shallow and slow/deep cognitive processing

Adrianna C. Jenkins, Rethinking Cognitive Load: A Default-Mode Network Perspective,Trends in Cognitive Sciences, Volume 23, Issue 7, 2019, Pages 531-533 DOI: 10.1016/j.tics.2019.04.008.

Typical cognitive load tasks are now known to deactivate the brain’s default-mode network (DMN). This raises the possibility that apparent effects of cognitive load could arise from disruptions of DMN processes, including social cognition. Cognitive load studies are reconsidered, with reinterpretations of past research and implications for dual-process theory.

A brief (and relatively shallow) account of computer programming as a cognitive ability

Evelina Fedorenko, Anna Ivanova, Riva Dhamala, Marina Umaschi Bers, The Language of Programming: A Cognitive Perspective, Trends in Cognitive Sciences,
Volume 23, Issue 7, 2019, Pages 525-528 DOI: 10.1016/j.tics.2019.04.010.

Computer programming is becoming essential across fields. Traditionally grouped with science, technology, engineering, and mathematics (STEM) disciplines, programming also bears parallels to natural languages. These parallels may translate into overlapping processing mechanisms. Investigating the cognitive basis of programming is important for understanding the human mind and could transform education practices.

A related work with a nice taxonomy of robot navigation algorithms

Eduardo J. Molinos, Ángel Llamazares, Manuel Ocaña Dynamic window based approaches for avoiding obstacles in moving, Robotics and Autonomous Systems,
Volume 118, 2019, Pages 112-130 DOI: 10.1016/j.robot.2019.05.003.

In recent years, Unmanned Ground Vehicles (UGVs) have been widely used as service robots. Unlike industrial robots, which are situated in fixed and controlled positions, UGVs work in dynamic environments, sharing the environment with other vehicles and humans. These robots should be able to move without colliding with any obstacle, assuring its integrity and the environment safety. In this paper, we propose two adaptations of the classical Dynamic Window algorithm for dealing with dynamic obstacles like Dynamic Window for Dynamic Obstacles (DW4DO) and Dynamic Window for Dynamic Obstacles Tree (DW4DOT). These new algorithms are compared with our previous algorithms based on Curvature Velocity Methods: Predicted Curvature Velocity Method (PCVM) and Dynamic Curvature Velocity Method (DCVM). Proposals have been validated in both simulated and real environment using several robotic platforms.

A Survey of Knowledge Representation in Service Robotics

avid Paulius, Yu Sun, A Survey of Knowledge Representation in Service Robotics,Robotics and Autonomous Systems, Volume 118, 2019, Pages 13-30 DOI: 10.1016/j.robot.2019.03.005.

Within the realm of service robotics, researchers have placed a great amount of effort into learning, understanding, and representing motions as manipulations for task execution by robots. The task of robot learning and problem-solving is very broad, as it integrates a variety of tasks such as object detection, activity recognition, task/motion planning, localization, knowledge representation and retrieval, and the intertwining of perception/vision and machine learning techniques. In this paper, we solely focus on knowledge representations and notably how knowledge is typically gathered, represented, and reproduced to solve problems as done by researchers in the past decades. In accordance with the definition of knowledge representations, we discuss the key distinction between such representations and useful learning models that have extensively been introduced and studied in recent years, such as machine learning, deep learning, probabilistic modeling, and semantic graphical structures. Along with an overview of such tools, we discuss the problems which have existed in robot learning and how they have been built and used as solutions, technologies or developments (if any) which have contributed to solving them. Finally, we discuss key principles that should be considered when designing an effective knowledge representation.