PALEO is an AI behavior layer for survival games. It interprets gameplay visuals, tracks dinosaur state, and recommends actions that feel instinctive and believable.
Teams use PALEO to validate map balance, improve encounter pacing, and prototype richer NPC behavior loops.
Reads gameplay frames and extracts behavior cues like threat, hunger, and opportunity.
Uses per-dinosaur memory so each creature keeps continuity, preferences, and risk history.
Outputs context-aware actions and explainable logs for design tuning and debugging.
Translates visual cues into predator/non-predator threat signals and safety-first action cues.
Computer Vision ResNet-18 trained OpenCV baselineEach dinosaur tracks identity, needs, recent events, threat memory, and explainable action intent.
Letta architecture State tracking Explainable logsFlags map imbalance, route failures, and survival-loop bottlenecks.
Developer tooling Stress testing Balancing feedbackSerengeti training, PoT fine-tuning, safety thresholding, deployment, and check-in deliverables complete.
Wire Letta as the real decision/memory layer in the live loop.
Run full in-game PoT smoke tests, add more holdout images, and collect demo clips.
Ambient map concept art with dense canopy layers, trails, and traversal chokepoints.
UI overlay concept showing interpreted tracks, age of trail, and direction confidence.
Panel concept for perception summary, risk score, memory recall, and selected action.
Spatial intensity map of predator sightings, prey density, and safe route zones.
Card set for hunger, thirst, stamina, threat memory, and recent decisions.
