Evidence for an Action Space - Motor Plans Activated Only for Objects Within Reach

AbstractThe digital curling game is a two-player zero-sum extensive game in a continuous action space. There are some challenging problems that are still not solved well, such as the uncertainty of strategy, the large game tree searching, and the use of large amounts of supervised data, etc. In this work, we combine NFSP and KR-UCT for digital curling games, where NFSP uses two adversary learning networks and can automatically produce supervised data, and KR-UCT can be used for large game tree searching in continuous action space. We propose two reward mechanisms to make reinforcement learning converge quickly. Experimental results validate the proposed method, and show the strategy model can reach the Nash equilibrium.

Download Full-text

Stationary Equilibria of Mean Field Games with Finite State and Action Space

Dynamic Games and Applications ◽

10.1007/s13235-019-00345-9 ◽

2020 ◽

Vol 10 (4) ◽

pp. 845-871 ◽

Cited By ~ 1

Author(s):

Berenice Anne Neumann

Keyword(s):

Mean Field ◽

Action Space ◽

Mean Field Games ◽

Finite State

Download Full-text

Perceiving actions before they happen: psychological dimensions scaffold neural action prediction

Social Cognitive and Affective Neuroscience ◽

10.1093/scan/nsaa126 ◽

2020 ◽

Author(s):

Mark A Thornton ◽

Diana I Tamir

Keyword(s):

Magnetic Resonance Imaging ◽

Functional Neuroimaging ◽

Social World ◽

Action Space ◽

Action Prediction ◽

Functional Magnetic Resonance ◽

Resonance Imaging ◽

The Social ◽

Psychological Dimensions ◽

The Brain

Abstract The social world buzzes with action. People constantly walk, talk, eat, work, play, snooze and so on. To interact with others successfully, we need to both understand their current actions and predict their future actions. Here we used functional neuroimaging to test the hypothesis that people do both at the same time: when the brain perceives an action, it simultaneously encodes likely future actions. Specifically, we hypothesized that the brain represents perceived actions using a map that encodes which actions will occur next: the six-dimensional Abstraction, Creation, Tradition, Food(-relevance), Animacy and Spiritualism Taxonomy (ACT-FAST) action space. Within this space, the closer two actions are, the more likely they are to precede or follow each other. To test this hypothesis, participants watched a video featuring naturalistic sequences of actions while undergoing functional magnetic resonance imaging (fMRI) scanning. We first use a decoding model to demonstrate that the brain uses ACT-FAST to represent current actions. We then successfully predicted as-yet unseen actions, up to three actions into the future, based on their proximity to the current action’s coordinates in ACT-FAST space. This finding suggests that the brain represents actions using a six-dimensional action space that gives people an automatic glimpse of future actions.

Download Full-text

Estimating Distances in Action Space in Augmented Reality

ACM Transactions on Applied Perception ◽

10.1145/3449067 ◽

2021 ◽

Vol 18 (2) ◽

pp. 1-16

Author(s):

Holly C. Gagnon ◽

Carlos Salas Rosales ◽

Ryan Mileris ◽

Jeanine K. Stefanucci ◽

Sarah H. Creem-Regehr ◽

...

Keyword(s):

Augmented Reality ◽

Real World ◽

Distance Measure ◽

Distance Perception ◽

Action Space ◽

Indoor Environments ◽

Verbal Reports ◽

Visual Matching ◽

Microsoft Hololens ◽

Real Person

Augmented reality ( AR ) is important for training complex tasks, such as navigation, assembly, and medical procedures. The effectiveness of such training may depend on accurate spatial localization of AR objects in the environment. This article presents two experiments that test egocentric distance perception in augmented reality within and at the boundaries of action space (up to 35 m) in comparison with distance perception in a matched real-world ( RW ) environment. Using the Microsoft HoloLens, in Experiment 1, participants in two different RW settings judged egocentric distances (ranging from 10 to 35 m) to an AR avatar or a real person using a visual matching measure. Distances to augmented targets were underestimated compared to real targets in the two indoor, RW contexts. Experiment 2 aimed to generalize the results to an absolute distance measure using verbal reports in one of the indoor environments. Similar to Experiment 1, distances to augmented targets were underestimated compared to real targets. We discuss these findings with respect to the importance of methodologies that directly compare performance in real and mediated environments, as well as the inherent differences present in mediated environments that are “matched” to the real world.

Download Full-text

On-Demand Channel Bonding in Heterogeneous WLANs: A Multi-Agent Deep Reinforcement Learning Approach

Sensors ◽

10.3390/s20102789 ◽

2020 ◽

Vol 20 (10) ◽

pp. 2789 ◽

Cited By ~ 1

Author(s):

Hang Qi ◽

Hao Huang ◽

Zhiqun Hu ◽

Xiangming Wen ◽

Zhaoming Lu

Keyword(s):

Reinforcement Learning ◽

Transmission Rate ◽

Single Agent ◽

Time Of Day ◽

Action Space ◽

Traffic Load ◽

Traffic Demand ◽

Channel Bonding ◽

On Demand ◽

Multi Agent

In order to meet the ever-increasing traffic demand of Wireless Local Area Networks (WLANs), channel bonding is introduced in IEEE 802.11 standards. Although channel bonding effectively increases the transmission rate, the wider channel reduces the number of non-overlapping channels and is more susceptible to interference. Meanwhile, the traffic load differs from one access point (AP) to another and changes significantly depending on the time of day. Therefore, the primary channel and channel bonding bandwidth should be carefully selected to meet traffic demand and guarantee the performance gain. In this paper, we proposed an On-Demand Channel Bonding (O-DCB) algorithm based on Deep Reinforcement Learning (DRL) for heterogeneous WLANs to reduce transmission delay, where the APs have different channel bonding capabilities. In this problem, the state space is continuous and the action space is discrete. However, the size of action space increases exponentially with the number of APs by using single-agent DRL, which severely affects the learning rate. To accelerate learning, Multi-Agent Deep Deterministic Policy Gradient (MADDPG) is used to train O-DCB. Real traffic traces collected from a campus WLAN are used to train and test O-DCB. Simulation results reveal that the proposed algorithm has good convergence and lower delay than other algorithms.

Download Full-text