Skip to main content
Formal Specification

Blind Granny (Polish Blind Man's Buff) — Formal Specification

Poland · chase-tag · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: discretizedGap-fills: 3
# Ciuciubabka — AI-optimized specification ## General idea An asymmetric hidden-information chase-and-tag game on a bounded 2D grid. The Chaser agent possesses zero direct visual sight and must reconstruct a probabilistic belief state over opponent coordinates from periodic noisy acoustic emissions (taunts, footsteps), execute a physical capture in proximity, and correctly infer the target's identity to pass on the blindfold role. ## Formal model - **Abstraction taken:** The continuous room is discretized into an 8×8 bounded Cartesian grid. The blindfold condition is modeled by setting the Chaser's visual observation tensor to null (∅). Instead, on each tick where an Evader moves or taunts, the Chaser receives an acoustic vector Z=distance+𝒩(0,σsound). The physical touch identification is modeled as a Bayesian classification over agent biometric features. - **Player/agent count:** 4 agents: 1 Blindfolded Chaser (C), 3 Evaders (E1,E2,E3). - **Turn model:** Simultaneous tick-based movement. All 4 agents submit action vectors simultaneously on each tick; actions resolve concurrently. ## Formal rules R1. **Setup:** Arena is an 8×8 grid. C starts at (4,4) with heading disorientation (random initial orientation). Evaders start at perimeter boundary cells. R2. **Action submission per tick:** - Evaders choose: `move(dx, dy)` (Δx,Δy∈{−1,0,1}), `taunt()`, or `freeze()`. - Chaser chooses: `move(dx, dy)` (Δx,Δy∈{−1,0,1}) or `lunge_grab(direction)`. R3. **Acoustic observation generation:** - If an Evader executes `taunt()`: emits sound with intensity I=1.0. - If an Evader executes `move()`: emits sound with intensity I=0.5. - If an Evader executes `freeze()`: emits sound with intensity I=0.0. - Chaser receives noisy bearing/distance signal yt=(d(C,Ei)+ϵ,θ(C,Ei)+δ). R4. **Capture check:** - If C executes `lunge_grab()` and an Evader Ek occupies the target cell or adjacent cell: Ek is captured. R5. **Identification phase (*Rozpoznanie*):** - Upon capture of Ek, C must submit `identify_captive(guessed_agent_id)`. - C observes a tactile feature vector F=true_feature(Ek)+𝒩(0,σtouch). - If guessed_agent_id==Ek: Round ends successfully; Ek becomes the Chaser for the next round. - If guessed_agent_id≠Ek: Capture is invalid; Ek is teleported 2 cells away, and C resumes hunting. R6. [gap-fill] **Round limit and victory:** A match consists of 3 rounds. The player who spends the least cumulative ticks in the Chaser role across all rounds wins the match. ## Edge cases - **Tie at the end condition:** If two agents accumulate identical chaser tick durations, [gap-fill] a 1v1 sudden-death 20-tick round determines the winner. - **A player with no legal move:** Boundary collisions restrict motion to interior directions; stationary `freeze()` is always legal. - **Repetition / cycles:** Bounded because C's belief distribution concentrates over time when evaders move/taunt, and the bounded 8×8 grid ensures high capture collision probability within Tmax⁡=100 ticks. - **Simultaneous claims on the same resource or position:** Multiple evaders cannot occupy the same grid cell; conflict resolves by higher agent ID standing down. - **Player elimination mid-round; turn order repair:** No elimination; roles swap dynamically upon successful capture-and-identification. - **Resource exhaustion:** Not applicable; grid and agents are persistent. - **Termination guarantee:** Guaranteed by Tmax⁡=100 tick timeout per round [gap-fill], after which a forced identification guess is triggered. ## State representation - **Persistent state:** `round_number` (1..3), `chaser_id` (AgentId), `cumulative_chaser_ticks` (Map[AgentId, Int]). - **Per-tick state:** `agent_coords` (Map[AgentId, (Int, Int)]), `captured_agent` (Option[AgentId]), `tick_count` (0..100). - **Observable information:** - Evaders observe full ground truth (positions of all agents). - Chaser observes only internal coordinate (xC,yC) and incoming acoustic emission signals yt. ## Action space - `move(dx: Int[-1..1], dy: Int[-1..1])`: Step to an adjacent cell. - `taunt()`: Evader only; emits loud directional audio cue. - `freeze()`: Evader only; suppresses movement noise. - `lunge_grab(direction: CompassDir)`: Chaser only; attempts physical grab on an adjacent cell. - `identify_captive(target_id: AgentId)`: Chaser post-grab action to guess captive's identity. ## Playability notes - **Single-player-vs-AI:** Highly compelling as a blind sensory stealth simulation (human as blindfolded hunter tracking AI sound signatures, or human as silent evader). - **Group games (n>4):** Reduces to 1 Chaser vs. N−1 Evaders natively. - **Physical-skill games:** Sound perception and tactile identification modeled via Bayesian filtering with tunable noise covariance matrices. - **What is lost:** Real-world tactile fabric textures, physical body heat, and vocal pitch mimicry. ## Implementation hints Use a simple particle filter or Kalman filter for the Chaser AI to track evader probability densities on the 8×8 grid. First milestone: implement the 2D grid with full visibility, then mask the Chaser's observation into the noisy acoustic sensor model.

Download as JSON

Download the full game corpus entry for this game as structured JSON.

Download JSON
Cite this page

Blind Granny (Polish Blind Man's Buff) (Poland). (n.d.). Games of Countries. https://gamesofcountries.com/games/pl-poland/ciuciubabka

@misc{gamesofcountries-pl-poland-ciuciubabka,
  title = {Blind Granny (Polish Blind Man's Buff) (Poland)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/pl-poland/ciuciubabka}
}