Research in the field of social robotics is allowing service robots to operate in environments with people. In the aim of realizing the vision of humans and robots coexisting in the same environment, several solutions have been proposed to (1) perceive persons and objects in the immediate environment; (2) predict the movements of humans; as well as (3) plan the navigation in agreement with socially accepted rules. In this work, we discuss the different aspects related to social navigation in the context of our experience in an indoor environment. We describe state-of-the-art approaches and experiment with existing methods to analyze their performance in practice. From this study, we gather first-hand insights into the limitations of current solutions and identify possible research directions to address the open challenges. In particular, this paper focuses on topics related to perception at the hardware and application levels, including 2D and 3D sensors, geometric and mainly semantic mapping, the prediction of people trajectories (physics-, pattern- and planning-based), and social navigation (reactive and predictive) in indoor environments.
An implementation of a parallel version of the shortest path algorithm on a Virtex-II Pro FPGA device that computes the minimal distance in a graph in a more efficiently way than the classical algorithms is presented The paper shows how the hardware/software codesign process is applied in order to design the system using a PowerPC processor running Linux on a XUP Virtex-II Pro development board. The coprocessor's hardware architecture is fully described as well as the software running in Linux that is in charge of transferring data between the host computer, the PPC and the application-specific coprocessor. The synthesis results are presented as well as a comparative study of speedups for the parallel and the sequential implementation of the algorithm, showing a good improvement from the presented version against a software version running in a PC.
The advancements in the robotic field have made it possible for service robots to increasingly become part of everyday indoor scenarios. Their ability to operate and reach defined goals depends on the perception and understanding of their surrounding environment. Detecting and positioning objects as well as people in an accurate semantic map are, therefore, essential tasks that a robot needs to carry out. In this work, we walk an alternative path to build semantic maps of indoor scenarios. Instead of relying on high-density sensory input, like the one provided by an RGB-D camera, and resource-intensive processing algorithms, like the ones based on deep learning, we investigate the use of low-density point-clouds provided by 3D LiDARs together with a set of practical segmentation methods for the detection of objects. By focusing on the physical structure of the objects of interest, it is possible to remove complex training phases and exploit sensors with lower resolution but wider Field of View (FoV). Our evaluation shows that our approach can achieve comparable (if not better) performance in object labeling and positioning with a significant decrease in processing time than established approaches based on deep learning methods. As a side-effect of using low-density point-clouds, we also better support people privacy as the lower resolution inherently prevents the use of techniques like face recognition.
This paper proposes Point-cloud Fast Filter (PFF), an algorithm to process efficiently 3D data from point-cloud sensors in order to build traversability maps for service robots. Our method is intended to be integrated with a 2D mapping algorithm, enhancing 2D standard maps with enough traversability information for robot navigation in indoor structured environments. The method is agnostic to the 3D sensor or mapping algorithm used, and keeps computational requirements low. Thus, we enable middle-class computers and a wide variety of sensors to be employed for service robots, reducing the costs of the platform. We evaluate the performance of PFF with different 3D sensors on a real robot and its impact on mapping, comparing it with alternative 2D and 3D mapping approaches.
In order to acquire socially acceptable behaviors and seamlessly interact with humans, robots must employ robust solutions to identify, track and follow the interacting person in common environments. Robots nowadays typically are equipped with a camera which can be used to perform body as well as face recognition. However, as the person moves in the environment, the robot might lose track of the individual, e.g., behind corners or obstacles. We report our experience in the design and implementation of a solution that is able to track and follow a target person and that can re-identify the target if the line of sight is obstructed temporarily during following. Our initial experimentation highlights the impact of the employed sensor and the environment in which the person moves on the accuracy with which a persons can be re-identified in practice.
Enabling robots to effectively detect people in their surrounding is essential to support operation in social environments. Due to the variety of robotic platforms and sensors available, as well as the constraints imposed on the use of cameras in some scenarios, it is essential to identify solutions able to work across different perception technologies. At the same time, solutions should not impose strong constraints on computing and memory resources, thus allowing robots to perform also complex tasks without affecting their reactiveness. To achieve such a goal, this article introduces Point-Cloud Fast Filter for People Detection (PFF- PED). The algorithm detects people by processing effectively point-cloud data, independently of the specific 3D sensor employed. By focusing on the most informative observations, PFF-PED is able to improve both accuracy and processing times compared to existing 2D and 3D solutions in our initial experimentation in indoor scenarios.