Search

Cooperative spatial exploration in initially unknown surroundings is a common embodied task in various applications and requires satisfactory coordination among the agents. Unlike many other research questions, there is a lack of simulation platforms for the cooperative exploration problem to perform and statistically evaluate different methods before they are deployed in practical scenarios. To this end, this paper designs a simulation framework to run different models, which features efficient event scheduling and data sharing. On top of such a framework, we propose and implement two different cooperative exploration strategies, i.e., the synchronous and asynchronous ones. While the coordination in the former approach is conducted after gathering the perceptive information from all agents in each round, the latter enables an ad-hoc coordination. Accordingly, they exploit different principles for assigning target points for the agents. Extensive experiments on different types of environments and settings not only validate the scheduling efficiency of our simulation engine, but also demonstrate the respective advantages of the two strategies on different metrics.

In distributed machine learning training, bulk synchronous parallel (BSP) and asynchronous parallel (ASP) are two main synchronization methods to help achieve gradient aggregation. However, BSP needs longer training time due to "stragglers" problem, while ASP sacrifices the accuracy due to "gradient staleness" problem. In this article, we propose a distributed training paradigm on parameter server framework called adaptive synchronous strategy (A2S) which improves the BSP and ASP paradigms by adaptively adopting different parallel training schemes for workers with different training speeds. Based on the stale value between the fastest and slowest worker, A2S adaptively adds a relaxed synchronous barrier for fast workers to alleviate gradient staleness, where a differentiated weighting gradient aggregation method is used to reduce the impact of slow gradients. Simultaneously, A2S adopts ASP training for slow workers to eliminate stragglers. Hence, A2S not only improves the "gradient staleness" and "stragglers" problems, but also obtains convergence stability and synchronous gain through synchronous and asynchronous parallel, respectively. Specially, we theoretically proved the convergence of A2S by deriving the regret bound. Moreover, experiment results show that A2S improves accuracy by up to 2.64% and accelerates training by up to 41% more than the state‐of‐the‐art synchronization methods BSP, ASP, stale synchronous parallel (SSP), dynamic SSP, and Sync‐switch. [ABSTRACT FROM AUTHOR]

If two people row a boat, they often call to each other to synchronize their strokes. It is anticipated that such a call promotes periodic joint action. The present study thus examined the effects of speech on both complementary and synchronous strategies in joint action using the same task as we used previously (Masumoto and Inui in J Neurophysiol 109:1307-1314, ). Ten pairs of participants produced periodic isometric forces such that the sum of the forces they produced was the target force cycling between 5 and 10 % of maximum voluntary contraction with an interval of 1,000 ms with the right hand. There were three speech conditions crossed with the presence or absence of visual information. Whereas two participants synchronized an utterance/ba/with the peak and valley forces in the 'Both' condition, one synchronized it with both forces in the 'One-side' condition, and nobody uttered it in the 'None' condition. When the total force was visible, the One-side and Both conditions exhibited lower correlations than the None condition, although the correlation between forces produced by two participants was negative in all conditions. When the total force was invisible, although the coherence between force and time series produced by two participants was low under the None condition, it was high at 1 and 3 Hz under the One-side and Both conditions. Thus, although periodically uttering a syllable worsened complementary force production when the target was visible, it promoted synchronization of their performance to each other's timing when the target was invisible. [ABSTRACT FROM AUTHOR]