Exploiting environmental signals to enable policy correlation in large-scale decentralized systems
نویسندگان
چکیده
Can artificial agents benefit from human conventions? Human societies manage to successfully self-organize and resolve the tragedy of commons in common-pool resources, spite bleak prediction non-cooperative game theory. On top that, real-world problems are inherently large-scale low observability. One key concept that facilitates coordination such settings is use conventions. Inspired by behavior, we investigate learning dynamics emergence temporal conventions, focusing on resources. Extra emphasis was given designing a realistic evaluation setting: (a) environment modeled fisheries, (b) assume decentralized learning, where can observe only their own history, (c) run simulations (up 64 agents). Uncoupled policies observability make cooperation hard achieve; as number grow, probability taking correct gradient direction decreases exponentially. By introducing an arbitrary common signal (e.g., date, time, or any periodic set numbers) means couple process, show conventions emerge reach sustainable harvesting strategies. The introduction consistently improves social welfare (by 258% average, up 3306%), range environmental parameters sustainability be achieved 46% 300%), convergence speed abundance 13% 53%).
منابع مشابه
Decentralized Adaptive Control of Large-Scale Non-affine Nonlinear Time-Delay Systems Using Neural Networks
In this paper, a decentralized adaptive neural controller is proposed for a class of large-scale nonlinear systems with unknown nonlinear, non-affine subsystems and unknown nonlinear time-delay interconnections. The stability of the closed loop system is guaranteed through Lyapunov-Krasovskii stability analysis. Simulation results are provided to show the effectiveness of the proposed approache...
متن کاملOptimal decentralized control of large scale systems
This paper presents a new optimized decentralized controller designmethod for solving the tracking and disturbance rejection problems for large-scale linear time-invariant systems, using only low-order decentralized controllers. To illustrate the type of results which can be obtained using the new optimized decentralized control design method, the control of a large flexible space structure is ...
متن کاملHybrid Decentralized Control of Large Scale Systems
Motivated by three applications which are under investigation at the Honeywell Research Laboratory in Minneapolis, we introduce a class of large scale control problems. In particular we show that a formation flight problem, a paper machine control problem and the coordination of cameras in a monitoring network can be cast into this class. In the second part of the paper we propose a decentraliz...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Autonomous Agents and Multi-Agent Systems
سال: 2022
ISSN: ['1387-2532', '1573-7454']
DOI: https://doi.org/10.1007/s10458-021-09541-7