Skip to content

Adaptive Intrinsic Motivation

Status: TRL-4 (integrated into every swarm node)

The Adaptive Intrinsic Motivation layer replaces static weights between survival, capital, and curiosity with a Meta‑POMDP agent that dynamically selects one of five macroscenarios:

  • safe_expansion – balanced trade / hide / research.
  • active_hunting – aggressive trading when DQ is low and capital sufficient.
  • stealth_mode – priority on reducing DQ, minimal trading.
  • exploration – high mutation rate and reduced curiosity threshold to discover new strategies.
  • crisis – survival at all costs, trading virtually stopped.

Formal verification

AdaptiveMotivation.tla proves that the sum of weights always equals 1 and that survival dominates in crisis.

Integration

The agent updates weights every 100 market steps, adjusting SurvivalEvaluator λ, hide aggressiveness, CuriosityEngine threshold, and GeneticEngine.mutation_rate. It also feeds into the speciation mechanism (niche selection).