Graph Neural Networks for Human-aware Social Navigation

Abstract

Autonomous navigation is a key skill for assistive and service robots. To be successful, robots have to navigate avoiding going through the personal spaces of the people surrounding them. Complying with social rules such as not getting in the middle of human-to-human and human-to-object interactions is also important. This paper suggests using Graph Neural Networks to model how inconvenient the presence of a robot would be in a particular scenario according to learned human conventions so that it can be used by path planning algorithms. To do so, we propose two ways of modelling social interactions using graphs and benchmark them with different Graph Neural Networks using the SocNav1 dataset. We achieve close-to-human performance in the dataset and argue that, in addition to promising results, the main advantage of the approach is its scalability in terms of the number of social factors that can be considered and easily embedded in code, in comparison with model-based approaches. The code used to train and test the resulting graph neural network is available in a public repository.

A survey on socially aware robot navigation: Taxonomy and future challenges

“…Other proposals consider environment information, although not of semantic type (e.g. size, structure), for navigation (Vega-Magro et al 2018;Manso et al 2019).…”

Generation of Human-Aware Navigation Maps Using Graph Neural Networks

“…A dataset containing scalars as output data cannot directly be used to train a model which provides two dimensional output, so the approach followed in this case is to use a model which provides one-dimensional value estimations (SNGNN-1D [24]) and sample its output shifting the robot's position, bootstrapping this way a two-dimensional dataset. The process of sampling is depicted in Fig.…”

“…Considering that the input data is not presented in the form of a graph, its conversion to a graph-like structure is one of the most relevant steps if GNNs are to be used. This process follows the same steps as [24], with the exception that there is an additional grid of 18x18 nodes whose values are passed to the CNN layers of the architecture and decoded into the final output. The first part of the graph, which coincides with [24] represents the entities in the room and their relations, using a node per entity (room, humans, walls and objects).…”

“…This process follows the same steps as [24], with the exception that there is an additional grid of 18x18 nodes whose values are passed to the CNN layers of the architecture and decoded into the final output. The first part of the graph, which coincides with [24] represents the entities in the room and their relations, using a node per entity (room, humans, walls and objects). The walls are split into segments, creating a node for each of these segments.…”

Multi-camera Torso Pose Estimation using Graph Neural Networks

“…A common application of human localisation and orientation is predicting intentions and movements in surveillance video feeds [1], [2]. An accurate localisation and orientation estimation are also crucial for human-aware navigation [3]. For instance, the orientation of pedestrians' velocity vectors is used in [4] to make a robot navigate in crowded environments complying with constraints defined by proxemics.…”