Ethical decision-making is difficult, certainly for robots let alone humans. If a robot's ethical decisionmaking process is going to be designed based on some approximation of how humans operate, then the assumption is that a good model of how humans make an ethical choice is readily available. Yet no single ethical framework seems sufficient to capture the diversity of human ethical decision making. Our work seeks to develop the computational underpinnings that will allow a robot to use multiple ethical frameworks that guide it towards doing the right thing. As a step towards this goal, we have collected data investigating how regular adults and ethics experts approach ethical decisions related to the use of deception in a healthcare and game playing scenario. The decisions made by the former group is intended to represent an approximation of a folk morality approach to these dilemmas. On the other hand, experts were asked to judge what decision would result if a person was using one of several different types of ethical frameworks. The resulting data may reveal which features of the pill sorting and game playing scenarios contribute to similarities and differences between expert and non-expert responses. This type of approach to programming a robot may one day be able to rely on specific features of an interaction to determine which ethical framework to use in the robot's decision making.
Recognising intent in collaborative human robot tasks can improve team performance and human perception of robots. Intent can differ from the observed outcome in the presence of mistakes which are likely in physically dynamic tasks. We created a dataset of 1227 throws of a ball at a target from 10 participants and observed that 47% of throws were mistakes with 16% completely missing the target. Our research leverages facial images capturing the person’s reaction to the outcome of a throw to predict when the resulting throw is a mistake and then we determine the actual intent of the throw. The approach we propose for outcome prediction performs 38% better than the two-stream architecture used previously for this task on front-on videos. In addition, we propose a 1D-CNN model which is used in conjunction with priors learned from the frequency of mistakes to provide an end-to-end pipeline for outcome and intent recognition in this throwing task.
The Penn State Autonomous Robotics Competition Club has developed a platform tailored for use in Search and Rescue missions to support first responders. The key design drivers included the capacity to carry a 10 lb LTE system to extend coverage for first responders on the ground, capability for long endurance flights, cost effectiveness, and ease of setup and operation for a solo operator. A coaxial octocopter with a hybrid power configuration was designed, verified in simulation, and flight tested. With an estimated operational endurance of over 3 hours and high levels of autonomy including obstacle avoidance and object detection, the proposed solution provides a versatile configuration capable of numerous applications including emergency operations.
Currently, 3D Computational Fluid Dynamic (CFD) rotorcraft simulations are able to account for blade crossover interaction (BCI) events, which are impulsive loading events that have a large influence on the vibrations and acoustics of a vehicle. Unfortunately, lower fidelity models are unable to adequately predict the BCI events and 3D CFD simulations are computationally expensive. This paper proposes a surrogate model that is able to predict the BCI event with reasonable accuracy and high computational efficiency for a subset of operating conditions. A dataset was created using transient 2D CFD simulations of airfoils moving toward each other in close proximity. The Mach number, angle-of-attack of each airfoil, along with the vertical separation distance between airfoils was varied in each simulation. An additional dependent parametric input (airfoil horizontal separation distance) was recorded along with the transient loads on the airfoils. This dataset was used in a supervised manner to train univariate (UV) and multivariate (MV) implementations of the Gaussian Process Regression model. Given airfoil operating conditions, the models are able to predict the BCI events. Hyper-parameters such as kernels and trend functions were compared using 5-fold cross validation and the final MV and UV models were compared on a held out test set to demonstrate predictive performance on unseen data.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.