Gemini AI Companion Bots

[1.2.22] - 2026-10-30

Fixed

• Villager Colony Breeding Requirements & Autonomous Self-Breeding Prevention:
  • Fixed a logic bug where a single bot with food and a spare bed could reproduce by itself.
  • Enforced authentic two-parent pairing: breeding strictly requires two non-baby villager companions standing within 16 blocks of each other.
  • Added 50-block radius population density check: breeding only occurs if there are at least 2 adult villagers within a 50-block radius (countVillagersIn50Radius >= 2).
  • Added strict 90%+ hunger threshold: both villager companions must have a hunger level of at least 18/20 (foodLevel >= 18) before breeding hearts can activate.
  • Enforced bed surplus condition: total beds in the 50-block range must equal or exceed total bot population + 1 (totalBeds >= botCount + 1) with an unassigned, vacant bed available.
  • Breeding food (3 wheat, 3 bread, 12 carrots, 12 potatoes, or 12 beetroots) is now consumed from both partner inventories upon birth.

[1.2.21] - 2026-10-30

Added

• Vanilla Villager / Mob 3D Voxel A Pathfinder (BotPathfinder):*
  • Implemented a deterministic 3D Voxel A* pathfinder modeled directly after Minecraft's WalkNodeEvaluator and GroundPathNavigation as the base navigation layer, with HiveMindBrain learned trail pheromones and hazard penalties layered on top as edge-cost modifiers.
  • Treats closed wooden doors (DoorBlock) and closed fence gates (FenceGateBlock) as openable passageways, enforces strict cardinal alignment through door/gate passages, and forbids corner-clipping diagonal transitions when either adjacent cardinal block is solid.
  • Automatically routes bots around exterior house walls and through front doors to reach indoor beds and workstations instead of walking in a straight line into the outside wall or corner of a house.
• Repetitive-Loop / Insanity Guard & Unreachable Target Blacklisting:
  • Added real-time stagnation detection (insanityStuckTicks): if a bot fails to advance >1.2 blocks within 25 ticks while trying to move, it automatically penalizes the obstructed corner/wall in HiveMindBrain and forces a fresh A* path calculation. If a target remains unreachable after 90 ticks (or is inside a completely sealed building with no door), it is blacklisted for 30 seconds and the bot stops pushing against the wall.

Fixed

• Eliminated Erratic Ground Jumping:
  • Removed all ground-jumping triggers on horizontalCollision, staircases, neural exploration (ACT_JUMP_ASCEND), unstuck fallbacks, and baby villager play routines. Because bots have STEP_HEIGHT = 1.06, they step up 1-block ledges and stairs smoothly without jumping; jumping now only occurs when swimming in water/lava/fluids or crossing a verified 1-block pit gap.
  • Reset setJumping(false) every tick on dry land so jump state can never remain stuck on.
• Accurate Passive Task Telemetry & Immediate "stop jumping" / "stop" Response:
  • Updated getCurrentTaskSummary() to report active passive states (e.g., walking indoors to bed, sheltering, or moving autonomously) instead of reporting "Idle (waiting for orders)" during passive movement, preventing Gemini from falsely claiming the bot is standing still.
  • Added immediate handling for "stop jumping", "quit jumping", "use the door", and "stop" so bots immediately halt jumping/movement, set standingBy = true, and penalize the stuck spot in the Hive-Mind.

[1.2.20] - 2026-10-29

Fixed

• Doorway & Fence Gate Navigation & Shoulder Collision:
  • Resolved an issue where companion bots became stuck or wedged in doorways, open fence gates, and 1-block chokepoints and hopped continuously in place.
  • Strict Lateral-Only Alignment: Replaced bidirectional center attractor spring forces with pure lateral alignment along the doorway cross-axis. Centering forces now only align the bot to the doorway centerline without applying backward drag or halting forward walking momentum.
  • Passage Axis Corridor Alignment: When entering, standing inside, or clearing a doorway or fence gate, bots now orient and walk strictly parallel to the opening's orientation (getDoorwayPassageAxis) rather than turning prematurely toward off-axis/diagonal targets. This preserves their narrow 0.6-block profile through the 0.8125-block door clearance and prevents shoulder clipping against doorframes and fence posts.
  • Doorway Jump Suppression: Jumping is now strictly suppressed while the bot is inside or traversing doorway and fence gate bounding boxes, preventing head collisions with door lintels and repetitive jumping against 1.5-block high fence collision boundaries.
  • Straight-Through Unstuck Bypass Waypoints: Unstuck routine now projects bypass waypoints 2 blocks straight through the opening along the passage corridor rather than at the bot's angled facing yaw, preventing waypoints from landing inside adjacent solid walls or fence posts.
  • Auto-Close Clearance Buffer: Adjusted doorway and gate auto-close logic so doors and fence gates do not swing shut until the bot is completely clear of the passage threshold.

[1.2.19] - 2026-10-29

Added

• Shared Hive-Mind Neural Network & Reinforcement Learning (HiveMindBrain):
  • Implemented a pure-Java Multi-Layer Perceptron (24 sensory inputs -> 32 hidden -> 16 hidden -> 9 micro-steering actions) with LeakyReLU activations, tanh Q-outputs, Eligibility Traces (γ = 0.92), Experience Replay, and online backpropagation shared across all companion bots on the server.
  • Added a persistent 3D Spatial Pheromone & Waypoint Graph (DIMENSION_NAV_GRAPHS) that records successful block traversals, directed neighbor edges, jump requirements, and hazard/stuck zones across dimensions.
  • Added a Modded Block Walkability Knowledge Base (BLOCK_KNOWLEDGE) that dynamically learns walkabilityScore, centerAlignmentNeed (shoulder-centering for narrow 1-block stairs/bridges), and jumpAssistNeed per Block Registry ID (such as Macaw's stairs and modded furniture).
  • Added a Role & Task Preference Q-Table (TASK_Q_VALUES) that tracks reinforcement scores for role:actionType pairs and injects top learned preferences and Hive-Mind IQ telemetry into Gemini AI prompts.
• Player Verbal & Command Reinforcement (/bot reward & /bot punish):
  • Players can praise or correct bots naturally in chat ("good job", "good bot", "well done", "nice pathing" vs. "bad job", "bad bot", "wrong way", "bad path", "stop getting stuck") or via /bot reward [name|all] [amount] and /bot punish [name|all] [amount].
  • Positive feedback propagates +15.0 reward backward through the bot's recent 60-step eligibility trace, strengthening neural weights, boosting spatial trail pheromones, and playing villager celebration sounds + happy villager particles.
  • Negative feedback propagates -15.0 punishment across the recent trajectory, marking stuck coordinates as hazard zones, triggering immediate lateral rerouting, and playing villager headshake feedback + angry villager particles.
• Command-Enabled Player Footstep Imitation (/bot learnpath):
  • Added /bot learnpath [start|stop] (and natural chat triggers "learn this path", "watch how I walk", "stop learning path") allowing players to demonstrate tricky routes, stairs, and jumps. Recorded footsteps are stored as golden breadcrumb nodes (+35.0 score) in the shared Hive-Mind graph for all bots to follow.
• Hive-Mind Diagnostics & 3D Trail Visualizer (/bot brain):
  • Added /bot brain [stats|show|reset] to inspect Hive-Mind IQ, training updates, learned modded block walkability, and top Q-value tasks, or render 3D particles in-world (END_ROD for player-taught golden paths, HAPPY_VILLAGER for learned trails, and FLAME for avoided stuck/hazard zones).