A planar tracking strategy based on multiple-interpretable improved PPO algorithm with few-shot technique.

Facing to a planar tracking problem, a multiple-interpretable improved Proximal Policy Optimization (PPO) algorithm with few-shot technique is proposed, namely F-GBQ-PPO. Compared with the normal PPO, the main improvements of F-GBQ-PPO are to increase the interpretability, and reduce the consumption for real interaction samples. Considering to increase incomprehensibility of a tracking policy, three levels of interpretabilities has been studied, including the perceptual, logical and mathematical interpretabilities. Detailly speaking, it is realized through introducing a guided policy based on Apollonius circle, a hybrid exploration policy based on biological motions, and the update of external parameters based on quantum genetic algorithm. Besides, to deal with the potential lack of real interaction samples in real applications, a few-shot technique is contained in the algorithm, which mainly generate fake samples through a multi-dimension Gaussian process. By mixing fake samples with real ones in a certain proportion, the demand for real samples can be reduced.

UNMAS: Multiagent Reinforcement Learning for Unshaped Cooperative Scenarios.

Multiagent reinforcement learning methods, such as VDN, QMIX, and QTRAN, that adopt centralized training with decentralized execution (CTDE) framework have shown promising results in cooperation and competition. However, in some multiagent scenarios, the number of agents and the size of the action set actually vary over time. We call these unshaped scenarios, and the methods mentioned above fail in performing satisfyingly. In this article, we propose a new method, called Unshaped Networks for Multiagent Systems (UNMAS), which adapts to the number and size changes in multiagent systems. We propose the self-weighting mixing network to factorize the joint action-value. Its adaption to the change in agent number is attributed to the nonlinear mapping from each-agent Q value to the joint action-value with individual weights. Besides, in order to address the change in an action set, each agent constructs an individual action-value network that is composed of two streams to evaluate the constant environment-oriented subset and the varying unit-oriented subset. We evaluate UNMAS on various StarCraft II micromanagement scenarios and compare the results with several state-of-the-art MARL algorithms. The superiority of UNMAS is demonstrated by its highest winning rates especially on the most difficult scenario 3s5z_vs_3s6z. The agents learn to perform effectively cooperative behaviors, while other MARL algorithms fail. Animated demonstrations and source code are provided in https://sites.google.com/view/unmas.