Recently, we have been witnesses of accidents involving autonomous vehicles and their lack of sufficient information. One way to tackle this issue is to benefit from the perception of different view points, namely cooperative perception. We propose here a decentralized collaboration, i.e. peer-topeer, in which the agents are active in their quest for full perception by asking for specific areas in their surroundings on which they would like to know more. Ultimately, we want to optimize a trade-off between the maximization of knowledge about moving objects and the minimization of the total volume of information received from others, to limit communication costs and message processing time. For this, we propose a way to learn a communication policy that reverses the usual communication paradigm by only requesting from other vehicles what is unknown to the ego-vehicle, instead of filtering on the sender side. We tested three different generative models to be taken as base for a Deep Reinforcement Learning (DRL) algorithm, and compared them to a broadcasting policy and a policy randomly selecting areas. More precisely, we slightly modified a state-of-the-art generative model named Temporal Difference VAE (TD-VAE) to make it sequential. We named this variant Sequential TD-VAE (STD-VAE). We also proposed Locally Predictable VAE (LP-VAE), inspired by STD-VAE, designed to enhance its prediction capabilities. We showed that LP-VAE produced better belief states for prediction than STD-VAE, both as a standalone model and in the context of DRL. The last model we tested was a simple state-less model (Convolutional VAE). Experiments were conducted in the driving simulator CARLA, with vehicles exchanging parts of semantic grid maps. Policies learned based on LP-VAE featured the best trade-off, as long as future rewards were taken into account. Our best models reached on average a gain of 25% of the total complementary information, while only requesting about 5% of the ego-vehicle’s perceptual field. We also provided interpretable hyperparameters controlling the reward function, which makes this trade-off adjustable (e.g. allowing greater communication costs).