We deal with the problem of incorporating the most recent information in the system which may be crucial to update the system state, while maintaining resilience to faulty measurements. For this purpose, we propose and analyze a consensus protocol biased towards the latest information in the system. The convergence to consensus of the system is analyzed and two main approaches are introduced with stability ranges provided from sufficient conditions. A detailed analysis of weight-changing that achieves bias towards novelty in the system is provided. Numerical simulations demonstrate that final value and convergence rate of the system can be controlled. Finally, we provide system demonstration in a simple use-case example.

This paper presents a method for extending decision-making policies in active perception tasks for multi-agent systems within partially observable environments. Multiple agents obtain their policies by training in an environment of a certain size. Those policies are then used in the environments larger in size, that are divided into sub-environments of size similar (or smaller) to one that the agents were trained in. Learned policies are adapted accordingly by proposed Action-Space Reduced Policy (ASRP). By leveraging Multi-Agent Reinforcement Learning (MARL) within a POMDP framework, agents can use their learned policies across environments of differing complexity without requiring retraining. The consensus mechanism allows agents to maintain a common belief state, supporting collaborative decision-making based on observations from all agents involved. Validation of the method is conducted on the scenario of multi-agent exploration missions, demonstrating the use of extended policies and enhanced perception accuracy. Simulation results indicate expected success rates and decision-making times, regardless of the environment’s dimensionality. Potential applications for scalable, multi-agent perception systems are discussed, along with directions for future research.

Programming by Demonstration (PbD) is a simple and efficient way to program robots without coding. PbD enables unskilled operators to demonstrate and guide robots to execute even the most complex tasks. This work aims to compare two approaches to PbD with a comprehensive user study, focusing on a common human skill. Each participant had to demonstrate to a robot how to draw a simple pattern both using a virtual marker and kinesthetic teaching. To evaluate differences between these demonstration approaches, we conducted a user study with 24 participants, benchmarking programmed trajectories, NASA raw task load index (rTLX), and system usability scale (SUS). We evaluated the similarity of the executed trajectories measuring difference between demonstrated and ideal trajectories. Our results show that human demonstration using a virtual marker is on average 8 times faster, superior in terms of quality and imposes 2 times less overall workload than kinesthetic teaching.

Since the diagnosis of autism spectrum disorder (ASD) relies heavily on behavioral observations by experienced clinician, we seek to investigate whether parts of this job can be autonomously performed by a humanoid robot using only sensors available on-board. To that end, we developed a robot-assisted ASD diagnostic protocol. In this work we propose the Partially observable Markov decision process (POMDP) framework for such protocol which enables the robot to infer information about the state of the child based on observations of child's behavior. We extend our previous work by developing a protocol POMDP model which uses tasks of the protocol as actions. We devise a method to interface protocol and task models by using belief at the end of a task to generate observations for the protocol POMDP, resulting in a hierarchical POMDP framework. We evaluate our approach through an exploratory study with fifteen children (seven typically developing and eight with ASD).

This paper presents a functional imitation task aimed at facilitating Autism Spectrum Disorder (ASD) diagnostics in children. Imitation plays a key role in the development of social skills at a young age, and studies have shown that the ability to imitate is impaired in children with ASD. Therefore, we expect imitation-based tasks to have diagnostic value. In this paper, we introduce two novel elements of human-robot interaction in the context of autism diagnostics. Instead of pure motoric imitation, we propose imitation tasks involving real objects in the environment. The introduction of physical objects strongly emphasizes joint attention skills, another area that is typically impaired in children with ASD. Furthermore, we present simple object detection, manipulation, tracking and gesture recognition algorithms, suitable for real-time, onboard execution on the small-scale humanoid robot NAO. The proposed system paves the way for fully autonomous execution of diagnostic tasks, which would simplify the deployment of robotic assistants in clinical settings. The source code for all described functionalities has been made publicly available as open-source software. We present a preliminary evaluation of the proposed system with a control group of typically developing preschool children and a group of seven children diagnosed with ASD.