Uncertainty maximization in partially observable domains: A cognitive perspective

Faced with an ever-increasing complexity of their domains of application, artificial learning agents are now able to scale up in their ability to process an overwhelming amount of data. However, this comes at the cost of encoding and processing an increasing amount of redundant information. This work exploits the possibility of learning systems, applied in partially observable domains, to selectively focus on the specific type of information that is more likely related to the causal interaction among transitioning states. A temporal difference displacement criterion is defined to implement adaptive masking of the observations. It can enable a significant improvement of convergence of temporal difference algorithms applied to partially observable Markov processes, as shown by experiments performed under a variety of machine learning problems, ranging from highly complex visuals as Atari games to simple textbook control problems such as CartPole. The proposed framework can be added to most RL algorithms since it only affects the observation process, selecting the parts more promising to explain the dynamics of the environment and reducing the dimension of the observation space.

POLIMI-ITW-S: A large-scale dataset for human activity recognition in the wild

Human activity recognition is attracting increasing research attention. Many activity recognition datasets have been created to support the development and evaluation of new algorithms. Given the lack of datasets collected in real environments (In The Wild) to support human activity recognition in public spaces, we introduce a large-scale video dataset for activity recognition In The Wild: POLIMI-ITW-S. The fully labeled dataset consists of 22,161 RGB video clips (about 46 h) including 37 activity classes performed by 50 K+ subjects in real shopping malls. We evaluated the state-of-the-art models on this dataset and get relatively low accuracy. We release the dataset including the annotations composed by person tracking bounding boxes, 2-D skeleton, and activity labels for research use at: https://airlab.deib.polimi.it/polimi-itw-s-a-shopping-mall-dataset-in-the-wild.

Datasets for recognition of aggressive interactions of children toward robotic toys

The data is related to unwanted interactions between a person and a small robotic toy based on acceleration sensor embedded within the robotic toy. Three toys were considered namely, a stuffed panda, a stuffed robot, and an excavator. Each toy was embedded with an accelerometer to record the interactions. Five different unwanted interactions were performed by adult participants and children. The considered interactions were hit, shake, throw, pickup, drop, and idle for the no interaction case. The collected data contains the magnitude of the resultant acceleration from the interactions. The data was processed by extracting the instances of interactions. A secondary dataset was created from the original one by creating artificial sequences. This data article contains the processed data that can be used to explore different machine learning models and techniques in classifying such interactions. Online repository contains the files: https://doi.org/10.7910/DVN/FHOO0Q.

Communication in Human-Robot Interaction

PURPOSE OF REVIEW

To present the multi-faceted aspects of communication between robot and humans (HRI), putting in evidence that it is not limited to language-based interaction, but it includes all aspects that are relevant in communication among physical beings, exploiting all the available sensor channels.

RECENT FINDINGS

For specific purposes, machine learning algorithms could be exploited when data sets and appropriate algorithms are available.

SUMMARY

Together with linguistic aspects, physical aspects play an important role in HRI and make the difference with respect to the more limited human-computer interaction (HCI). A review of the recent literature about the exploitation of different interaction channels is presented. The interpretation of signals and the production of appropriate communication actions require to consider psychological, sociological, and practical aspects, which may affect the performance. Communication is just one of the functionalities of an interactive robot and, as all the others, will need to be benchmarked to support the possibility for social robots to reach a real market.

Data on the impact of an object with different thicknesses of different soft materials at different impact velocities on a dummy head

The purpose of this data is to investigate the effect of different thicknesses of different soft materials samples added to an object on the resultant head acceleration of a developed dummy head upon impact. The object was a cylinder (10 × 10 cm², diameter and height) and weighs 0.4 kg. The investigated materials were Ecoflex, Dragon Skin, and Clay while the thickness were 1 mm, 2 mm, 3 mm, and 5 mm. The velocities of the impacts for the 108 experiments were between 1 m/s and 3 m/s. Three severity indices (i.e. peak head linear acceleration, 3 ms criterion and the Head Injury Criterion (HIC)) were calculated from the raw acceleration data. The impact velocities were tabulated from the video recordings. A summary of the processed data and the raw data are included in this dataset. Online repository contains the files: https://doi.org/10.7910/DVN/TXOPUH.

Data on the impact of objects with different shapes, masses, and impact velocities on a dummy head

In this article, a data generated from impacts of objects with different shapes, masses, and impact velocities on a developed dummy head. The mass considered was in the range of 0.3-0.5 kg while the shapes considered were cube, wedge, and cylinder. The impact velocities levels were in the range of 1-3 m/s. A total of 144 experiments were conducted and the corresponding videos and raw data were analyzed for impact velocity, peak head linear acceleration, 3 ms criterion, and the Head Injury Criterion (HIC). This dataset includes the raw acceleration data and a summary of the overall processed data. The data is available on Harvard Dataverse: https://doi.org/10.7910/DVN/AVC8GG.

Functional evaluation and rehabilitation engineering

Life is complex and all about movement, which allows us to interact with the environment and communicate with each other. The human nervous system is capable of performing a simultaneous and integrated control of 100-150 mechanical degrees of freedom of movement in the body via tensions generated by about 700 muscles. In its widest context, movement is carried out by a sensory motor system comprising multiple sensors (visual,auditory, and proprioceptive),multiple actuators (muscles acting on the skeletal system),and an intermediary processor that can be summarized as a multiple-input–multiple-output nonlinear dynamic time-varying control system. This grand control system is capable of responding with remarkable accuracy,speed, appropriateness,versatility, and adaptability to a wide spectrum of continuous and discrete stimuli and conditions and is certainly orders of magnitude more complex and sophisticated than the most advanced robotic systems currently available. In the last decades,a great deal of research has been carried out in the fields of functional evaluation of human performance and rehabilitation engineering. These fields combine knowledge, concepts, and methods from across many disciplines (e.g., biomechanics,neuroscience, and physiology), with the aim of developing apparatuses and methods fort he measurement and analysis of complex sensory motor performance and the ultimate goal of enhancing the execution of different tasks in both healthy people and persons with reduced capabilities from different causes (injury, disease, amputation,and neural degeneration).

An approach to the design of reinforcement functions in real world, agent-based applications

The success of any reinforcement learning (RL) application is in large part due to the design of an appropriate reinforcement function. A methodological framework to support the design of reinforcement functions has not been defined yet, and this critical and often underestimated activity is left to the ability of the RL application designer. We propose an approach to support reinforcement function design in RL applications concerning learning behaviors for autonomous agents. We define some dimensions along which we can describe reinforcement functions; we consider the distribution of reinforcement values, their coherence and their matching with the designer's perspective. We give hints to define measures that objectively describe the reinforcement function; we discuss the trade-offs that should be considered to improve learning and we introduce the dimensions along which this improvement can be expected. The approach we are presenting is general enough to be adopted in a large number of RL projects. We show how to apply it in the design of learning classifier systems (LCS) applications. We consider a simple, but quite complete case study in evolutionary robotics, and we discuss reinforcement function design issues in this sample context.