Learning From Demonstration in the Wild

Abstract

Learning from demonstration (LfD) is useful in settings where hand-coding behaviour or a reward function is impractical. It has succeeded in a wide range of problems but typically relies on manually generated demonstrations or specially deployed sensors and has not generally been able to leverage the copious demonstrations available in the wild: those that capture behaviours that were occurring anyway using sensors that were already deployed for another purpose, e.g., traffic camera footage capturing demonstrations of natural behaviour of vehicles, cyclists, and pedestrians. We propose video to behaviour (ViBe), a new approach to learn models of behaviour from unlabelled raw video data of a traffic scene collected from a single, monocular, initially uncalibrated camera with ordinary resolution. Our approach calibrates the camera, detects relevant objects, tracks them through time, and uses the resulting trajectories to perform LfD, yielding models of naturalistic behaviour. We apply ViBe to raw videos of a traffic intersection and show that it can learn purely from videos, without additional expert knowledge.

Variable-drift diffusion models of pedestrian road-crossing decisions

“…Making use of models of pedestrian behaviour in automated vehicle algorithms is an active area of research, but so far the mod-els used have been relatively simplistic [9,29,30]. Another important role of human behavior models in vehicle development is as agents in simulation environments for virtual testing [1,9,41]. For both of these applications, since the current model only considers the specific case of a pedestrian who is stationary at the kerb, one would likely want to adapt the model into a larger framework for behavior prediction, to allow modeling of a richer variety of scenarios.…”

“…Existing approaches to computational modeling of road user behavior mirror the modeling paradigms in the wider cognitive and behavioral sciences, including cognitive architectures [56], ecological psychology [19], classical and optimal control theory [49], rational decision-making [11], game theory [16,28], as well as data-driven modeling using machine learning approaches [1,37]. However, most of these existing models have either emphasized detailed modeling of individual road user behavior, or more coarse-grained modeling of interactions of larger number of road users, for example to study high-level traffic flow.…”

SLAM – DATMO Method for an Autonomous Vehicles

“…sharing files between lots of thousand vehicles in a city need to no longer boost to be feasible. A semiotic framework that integrates specific sources of information and converts uncooked sensor information into considerable descriptions used to be delivered in [33] for this purpose, in [34], the time size. However, due to the excessive computational necessities and environmental challenges, on foot SLAM algorithms outdoors, which is the operational neighborhood of ADSs, is a entire lot a good deal much less environment fantastic than localization with a prebuilt map [12].…”

How Do We Study Pedestrian Interaction with Automated Vehicles? Preliminary Findings from the European interACT Project

“…Traffic microsimulation is a well-established field of research, and commercial software products exist that permit traffic simulations that are accurate on the scale of a large junction or a city centre, for example, to predict how a range of alternative road infrastructure designs will affect traffic throughput [23,24]. The road user behaviour models in these traffic simulations are, however, not designed to capture the details of local interactions, and this underdeveloped area is now garnering increasing attention, with some modellers approaching it from a traffic microsimulation starting point [26,27], and others addressing it as a data-driven machine learning challenge [28,29,30].…”