Perception as abduction: turning sensor data into meaningful representation

This article presents a formal theory of robot perception as a form of abduction. The theory pins down the process whereby low-level sensor data is transformed into a symbolic representation of the external world, drawing together aspects such as incompleteness, top-down information flow, active perception, attention, and sensor fusion in a unifying framework. In addition, a number of themes are identified that are common to both the engineer concerned with developing a rigorous theory of perception, such as the one on offer here, and the philosopher of mind who is exercised by questions relating to mental representation and intentionality.     

Neuro-cognitive mechanisms of decision making in joint action: a human-robot interaction study

In this paper we present a model for action preparation and decision making in cooperative tasks that is inspired by recent experimental findings about the neuro-cognitive mechanisms supporting joint action in humans. It implements the coordination of actions and goals among the partners as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of others’ motor behavior. The control architecture is formalized by a system of coupled dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode task-relevant information about action means, task goals and context in the form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic model of joint action is evaluated in a task in which a robot and a human jointly construct a toy object. We show that the highly context sensitive mapping from action observation onto appropriate complementary actions allows coping with dynamically changing joint action situations.     

The anticipatory construction of reality as a central concern for psychology and robotics

How is it that, given our diverse sensors that are moving at any moment, we get to the idea that there is a more or less permanent world around us that contains objects and living beings and that is endowed with spatial and temporal properties? This question is difficult, it is “the child’s construction of reality” issue. Given the complexity of the underlying phenomena, it requires a dialog between engineering sciences and life sciences to be solved. In this paper, we introduce the contributions to this dialog that have been collected in this special issue.     

A single sensor system for mapping in GNSS-denied environments

Autonomous robotic vehicles need accurate positioning to navigate. For outdoor autonomous vehicles, the localization problem has been solved using GNSS systems. However, many places suffer from problems in the signal of those systems, known as GNSS-denied environments. To face such a problem, several approaches first map the environment to thereafter localize the vehicle within it. The solutions for mapping in GNSS-denied places, mostly, are tested indoors, and the revisited areas are generally close from the starting position. In this work, we develop a single sensor system for mapping in large-scale GNSS-denied sites, based exclusively on a 3D LiDAR system. The proposed work consists of a 3D-LiDAR with a LiDAR Odometry approach (LO) estimating movements between frames thus providing of dead-reckoning estimates of the vehicle. The LiDAR Odometry is the input to a virtual GNSS system based on a Particle Filter Localization which matches the estimated dead-reckoning with a road map, with the assumption that the vehicle usually navigates on roads. The global position produced by the virtual GNSS is used to detect and correct loop closures in case of revisited areas. A GraphSLAM implementation fuses the outcomes from LiDAR Odometry, virtual GNSS and loop closure, and yield a feasible pose of the vehicle. Finally, a mapping procedure places every 3D frame and builds 2D occupancy grid maps. The system developed here is evaluated empirically using two datasets collected in a dynamic environment with paths of 3.7 and 6.5 km long, respectively. For both datasets, the system presents an RMS of 6.5 m according to GNSS sensor readings used for comparison purposes.     

Decision-making in robotics and psychology: A distributed account

Decision-making is usually a secondary topic in psychology, relegated to the last chapters of textbooks. The psychological study of decision-making assumes a certain conception of its nature and mechanisms that has been shown wrong by research in robotics. Robotics indicate that decision-making is not—or at least not only—an intellectual task, but also a process of dynamic behavioral control, mediated by embodied and situated sensorimotor interaction. The implications of this conception for psychology are discussed.     

Cognition as coordinated non-cognition

We propose that cognition is more than a collection of independent processes operating in a modular cognitive system. Instead, we propose that cognition emerges from dependencies between all of the basic systems in the brain, including goal management, perception, action, memory, reward, affect, and learning. Furthermore, human cognition reflects its social evolution and context, as well as contributions from a developmental process. After presenting these themes, we illustrate their application to the process of anticipation. Specifically, we propose that anticipations occur extensively across domains (i.e., goal management, perception, action, reward, affect, and learning) in coordinated manners. We also propose that anticipation is central to situated action and to social interaction, and that many of its key features reflect the process of development.     

Neural Policy Style Transfer

Style Transfer has been proposed in a number of fields: fine arts, natural language processing, and fixed trajectories. We scale this concept up to control policies within a Deep Reinforcement Learning infrastructure. Each network is trained to maximize the expected reward, which typically encodes the goal of an action, and can be described as the content. The expressive power of deep neural networks enables encoding a secondary task, which can be described as the style. The Neural Policy Style Transfer (NPST)¹ algorithm is proposed to transfer the style of one policy to another, while maintaining the content of the latter. Different policies are defined via Deep Q-Network architectures. These models are trained using demonstrations through Inverse Reinforcement Learning. Two different sets of user demonstrations are performed, one for content and other for style. Different styles are encoded as defined by user demonstrations. The generated policy is the result of feeding a content policy and a style policy to the NPST algorithm. Experiments are performed in a catch-ball game inspired by the Deep Reinforcement Learning classical Atari games; and a real-world painting scenario with a full-sized humanoid robot, based on previous works of the authors. The implementation of three different Q-Network architectures (Shallow, Deep and Deep Recurrent Q-Network) to encode the policies within the NPST framework is proposed and the results obtained in the experiments with each of these architectures compared.     

Spiritual robots: Religion and our scientific view of the natural world

Religion plays a powerful role in the formation of scientific theories. By comparing the goals and practice of robotics and artificial intelligence in the US and Japan, differences between the two countries can be traced to their religious environments. Christian expectations of cosmic purpose and hope for salvation in purified, unearthly bodies leads to US researchers' preference for artificial intelligence over humanoid robots, a desire to see cosmic meaning in the development of that intelligence, and salvation of human minds in virtual, non-biological bodies. In Japan, robots, which have been the subjects of ritual consecrations and religious transcendence, participate in a fundamental sanctity of the natural world. A positive outlook on being human promotes a preference for humanoid robots and a future in which robots serve human beings, who do not forsake their bodies for virtual lives. Divergent scientific strategies cannot be separated from the religious worlds of their practitioners.     

The sense of agency in human-human vs human-robot joint action

Kinesthesis pertains to the perception of moving body parts, while the sense of agency refers to the experience of controlling one’s action-effects. Based on previous work, we hypothesized that the sense of agency would decrease in joint action with a robot compared to a human partner. Pairs of participants were jointly manipulating two interconnected haptic devices enabling them to feel each other’s forces. Unbeknown to participants, their partner was sometimes replaced by a robot. The sense of agency was assessed using intentional binding, which refers to a contraction of perceived time between an action and its effect for intentional actions, and participants’ judgment of their contribution to joint action. Participants judged their contribution as higher when they were initiating action and when they were paired with the robot. By contrast, intentional binding occurred only with a human partner. This outcome supports the hypothesis that human-robot joint action hinders intentional binding.