SocNav1: A Dataset to Benchmark and Learn Social Navigation Conventions

Abstract

Datasets are essential to the development and evaluation of machine learning and artificial intelligence algorithms. As new tasks are addressed, new datasets are required. Training algorithms for human-aware navigation is an example of this need. Different factors make designing and gathering data for human-aware navigation datasets challenging. Firstly, the problem itself is subjective, different dataset contributors will very frequently disagree to some extent on their labels. Secondly, the number of variables to consider is undetermined culture-dependent. This paper presents SocNav1, a dataset for social navigation conventions. SocNav1 aims at evaluating the robots’ ability to assess the level of discomfort that their presence might generate among humans. The 9280 samples in SocNav1 seem to be enough for machine learning purposes given the relatively small size of the data structures describing the scenarios. Furthermore, SocNav1 is particularly well-suited to be used to benchmark non-Euclidean machine learning algorithms such as graph neural networks. This paper describes the proposed dataset and the method employed to gather the data. To provide a further understanding of the nature of the dataset, an analysis and validation of the collected data are also presented.

A graph neural network to model disruption in human-aware robot navigation

“…Although the comparison was favourable, it is not entirely fair as the approaches have slightly different goals. We are aware of other researchers currently working with the dataset used in this paper and SocNav1 [25], but there are no published works to compare with at the time of writing. Most approaches introduced in Section 1 deal with modelling human intimate, personal, social and interaction spaces instead of social inconvenience, which is a more general term.…”

“…SocNav1 [25], was designed to learn and benchmark estimation functions for social navigation conventions. SocNav2 -presented in this paper-has the same goal as its predecessor but unlike SocNav1, it considers the velocity and trajectory of the robots and the humans around them.…”

Generation of Human-Aware Navigation Maps Using Graph Neural Networks

“…Specifically, SNGNN-2D combines GNNs and CNNs to generate a cost map from a graph representing the different elements of a room as well as the relationships between them. An interesting feature of SNGNN-2D is that the model is trained using a map-like 2D dataset bootstrapped from a singlepoint 1D model developed in a previous work [23]. In such 1D model (SNGNN-1D), for any given graph describing a scenario, the network generates a single scalar estimation of how disruptive the robot is overall for the people in the scenario.…”

“…For each scenario in the bootstrapped dataset a matrix of 73x73 samples is generated. A total of 37131 scenarios were randomly generated following the same strategy of SocNav1 [23]. The dataset split for training, development and test is of 31191, 2970 and 2970 scenarios, respectively.…”

Data from Smartphones and Wearables

“…The database described in [1], SocNav1, contains a dataset for social navigation conventions. SocNav1 aims at evaluating the robots' ability to assess the level of discomfort that their presence might generate among humans, which could be used in the future by robot navigation systems to estimate path costs.…”