Collaborative hunting in artificial agents with deep reinforcement learning.

Area of Science:

• Behavioral Ecology
• Computational Neuroscience
• Evolutionary Biology

Background:

• Collaborative hunting was thought to require high-level cognition and large brains.
• Recent observations of collaborative hunting in smaller-brained vertebrates challenge this notion.

Purpose of the Study:

• To investigate if complex collaborative hunting strategies can emerge from simple decision-making processes.
• To explore the cognitive requirements for coordinated predator behavior.

Main Methods:

• Utilized computational multi-agent simulations.
• Employed deep reinforcement learning techniques.
• Modeled predator decision-making based on internal representations and prior experience.

Main Results:

• Demonstrated that sophisticated coordination can arise from simple, experience-based decision rules.
• Showed that predator coordination is robust against unpredictable prey behavior.
• Found that distance-dependent internal representations are key to emergent cooperation.

Conclusions:

• Collaborative hunting does not necessarily depend on advanced cognitive abilities.
• Simple decision-making mechanisms can explain complex social behaviors in predators.
• Findings offer insights into the evolution of sociality and cooperation in nature.