Skip to main content
Formal Specification

Blind Grandmother — Formal Specification

Bulgaria · guessing-hidden · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 4
# Slyapa baba — AI-optimized specification ## General idea A spatial hide-and-seek pursuit game under complete visual occlusion with acoustic tracking and biometric identification. The seeker navigates a 2D bounded grid based on discrete sound emissions generated by evasive agents, grabs an agent within contact radius, and performs a classification query over feature attributes to identify the captive. ## Formal model - **Abstraction taken:** Visual blindness is modeled by setting the seeker's visual perception field to null; seeker receives noisy acoustic vectors (θ,intensity) from sound events. Tactile/voice identification is modeled as sampling from a confusion matrix over agent biometric feature vectors. - **Player/agent count:** N agents (1 Seeker B, N−1 Evasive agents E1..EN−1, N≥4). - **Turn model:** Simultaneous-step spatial movement ticks on a bounded W×H grid. ## Formal rules R1. Setup: Arena is a discrete 2D grid of size W×H (e.g. 10×10). Seeker B starts at grid center (x0,y0). Evasive agents start at random grid locations pi∈Arena∖{(x0,y0)}. R2. Movement tick: At each tick t, all agents simultaneously submit movement actions Δp∈{(0,0),(±1,0),(0,±1)}. R3. Sound generation and sensing: - [gap-fill] Whenever an evasive agent moves, they generate a sound emission with probability Psound=0.6. If an agent stays still, Psound=0.1. - Seeker receives an auditory observation list of active sound sources with angular and distance noise: pˆsound=pi+𝒩(0,σ2). R4. Boundary penalty: Any evasive agent moving outside the grid bounds [0,W−1]×[0,H−1] is immediately designated as captured. R5. Capture condition: If distance dist(pB,pi)≤1.0 (Manhattan distance ≤1), agent i is captured (Acaught=i) and movement freezes. R6. Identification phase (Guessing): - Each agent possesses a visible feature profile vector Vi=[height,hair,voice_pitch,build]. - [gap-fill] Seeker observes a noisy tactile/auditory feature sample Vˆ=Vi+ϵsensory. - Seeker outputs a discrete prediction yˆ∈{1,…,N−1}. R7. Guess resolution: - If yˆ==Acaught, the guess is correct: Acaught becomes the new Seeker for the next round; Seeker scores +1 point. - If yˆ≠Acaught, the guess fails: Acaught is released, seeker receives 0 points, and evasive agents are granted a 2-tick immunity sprint to reposition [gap-fill]. R8. [gap-fill] Round limit: If Seeker fails 3 consecutive identifications or round exceeds Tmax=100 ticks, the role rotates to the evasive agent who accumulated the most noise emissions. ## Edge cases - Tie at the end condition: Points represent successful seeker identifications; ties broken by fewest ticks taken to achieve identification. - A player with no legal move: Not applicable — stay action (0,0) is always legal. - Repetition / cycles: Bounded: The finite grid dimension W×H, non-zero seeker speed, and Tmax tick limit guarantee that either capture occurs or round times out. - Simultaneous claims on the same resource or position: Evasive agents can occupy the same cell; if seeker enters a multi-occupant cell, the captive is sampled uniformly at random among occupants. - Player elimination mid-round; turn order repair: If an evasive agent disconnects, their entity is removed; if seeker disconnects, the nearest evasive agent becomes the new seeker. - Resource exhaustion: Not applicable — grid space and feature profiles are static and non-depletable. - Termination guarantee: Enforced by the Tmax hard round cutoff and finite identification attempts per capture. ## State representation - Persistent state: Feature profiles V1..N−1, cumulative identification scores, active seeker ID. - Per-tick ephemeral state: Seeker position pB, evasive positions p1..N−1, generated sound list, capture flag, active captive ID. - Observable information: Evasive agents observe full grid state; Seeker observes ONLY noisy acoustic signals and tactile features upon contact. ## Action space - `move(dx, dy)`: Step displacement in {−1,0,1}2. - `make_sound(intent)`: Optional taunt action increasing seeker observation certainty. - `predict_identity(agent_id)`: Precondition: Captive held, executed by Seeker, argument ∈{1..N−1}. ## Playability notes - Single-player-vs-AI: Highly compelling — Human plays as blind seeker relying on audio spatialization / radar pings and deductive feature elimination, or as evasive prey minimizing acoustic footprint. - Group games (n>4): Reduces naturally to 2 agents (1 Seeker vs 1 Evasive prey) where the prey tries to evade for T ticks while generating mandatory noise signals. - Physical-skill games: Tactile exploration is abstracted into feature vector estimation over discrete categories. - What is lost: The physical comedy of human touch, voice pitch modulation, and giggling is formalized into Bayesian feature inference. ## Implementation hints Branching factor is 5 for movement. The core engine is a standard grid-world POMDP (Partially Observable Markov Decision Process). Suggested first milestone: implement the 2D grid pursuit with full visibility, then overlay the acoustic sensing mask and feature identification table.

Download as JSON

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

Download JSON
Cite this page

Blind Grandmother (Bulgaria). (n.d.). Games of Countries. https://gamesofcountries.com/games/bg-bulgaria/slyapa-baba

@misc{gamesofcountries-bg-bulgaria-slyapa-baba,
  title = {Blind Grandmother (Bulgaria)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/bg-bulgaria/slyapa-baba}
}