Formal Specification
Blindfold Tag and Identification / Blind Bee — Formal Specification
India · guessing-hidden · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: discretizedGap-fills: 3
# Aankh Micholi / Kanamachi — AI-optimized specification
## General idea
Aankh Micholi (known in Bengal as *Kanamachi* and in Tamil Nadu as *Kannamoochi*) is a traditional Indian blindfold pursuit and identification game. One blindfolded seeker operates with zero visual field data, navigating entirely via localized acoustic cues (claps, footsteps, chants) and tactile proximity inside a bounded arena. Upon grasping an evasive player, the seeker must correctly infer the captive's true identity from a discrete tactile/vocal feature vector to transfer the seeking role.
## Formal model
- **Abstraction taken:** The 10m 10m physical play space is discretized into a 2D integer grid ( cells). Continuous human movement and audio-tactile perception are discretized into discrete time ticks . Visual perception is masked completely for the Seeker (), who instead receives directional acoustic emission vectors generated by nearby evasive agents ().
- **Player/agent count:** agents ( Seeker , Evasive players ), where .
- **Turn model:** Simultaneous-move discrete time ticks.
## Formal rules
R1. **Initial Setup:**
- Arena bounded by grid .
- One agent is initialized as the Seeker with status .
- Evasive agents are initialized at distinct grid coordinates within the boundary.
- Each agent possesses a static public feature profile drawn from discrete domains.
R2. **Tick Dynamics & Movement:**
- On each tick , all agents simultaneously choose a movement vector .
- Seeker moves at maximum speed cell/tick without visual knowledge of agent coordinates.
- Evasive agents may move at speed cell/tick.
- If an evasive agent moves outside the boundary, they incur an immediate out-of-bounds penalty and are designated as the new Seeker immediately.
R3. **Acoustic Cue Generation:**
- On each tick, active evasive agents generate an acoustic signal of magnitude (0: silent stealth, 1: footsteps/chanting, 2: clapping/taunting).
- [gap-fill] An evasive agent must generate at least a magnitude-1 acoustic signal once every 3 ticks to satisfy the traditional mandatory chanting requirement (*"Kanamachi bho-bho"*).
- The Seeker receives a noisy relative vector for signals above an auditory threshold.
R4. **Capture & Freeze:**
- A capture occurs when Seeker and an evasive agent occupy the same cell , or adjacent orthogonal cells, and executes action `tag()`.
- Upon successful capture, enters state and must remain motionless.
R5. **Tactile/Vocal Identification Phase:**
- Seeker executes action `probe_feature()` to observe a noisy sample of the captive's feature profile .
- Seeker submits a guess .
- If (correct identification): Seeker is relieved, agent becomes the new Seeker , and a new round begins.
- If (incorrect identification): Captive is released back into the arena, granted a 3-tick invulnerability buffer to retreat, and resumes hunting.
R6. **Round & Match Termination:** [gap-fill] A match consists of a fixed number of rounds (e.g. 5 rounds), or terminates when every participant has served as the Seeker at least once.
## Edge cases
- **Tie at the end condition:** Not applicable — roles are individual; game continues in a rolling cycle with a single seeker at any moment.
- **A player with no legal move:** All agents can always choose `stay()` .
- **Repetition / cycles:** The cycle of missed identifications is bounded by [gap-fill] a maximum of 3 failed guesses per capture (after 3 failed guesses, the captive's identity is revealed and the seeker is replaced to prevent infinite entrapment).
- **Simultaneous claims on the same resource or position:** If the seeker is adjacent to multiple evasive agents upon `tag()`, the nearest agent with the lowest Euclidean distance is tagged; if tied, resolution chooses the lowest agent ID.
- **Player elimination mid-round; turn order repair:** Out-of-bounds infraction immediately triggers role transfer: the offending player becomes the new Seeker without requiring a tag/identification sequence.
- **Resource exhaustion (pieces, tokens, space, time):** Spatial boundaries are invariant; acoustic signal channels reset per tick.
- **Termination guarantee: why the game provably ends:** A single round terminates upon a correct identification or an out-of-bounds infraction; match termination is guaranteed by the fixed round limit .
## State representation
- **Persistent state:**
- : total agent count.
- : active seeker index.
- : ground-truth physical/vocal feature profiles for all agents.
- : completed rounds counter.
- **Per-tick ephemeral state:**
- Coordinates for all .
- Active acoustic emissions .
- .
- : ticks since last acoustic chant per evasive agent.
- **Observable information:**
- Evasive agents observe: Full visual state (positions of all agents, boundaries, seeker status).
- Seeker observes: No visual coordinate data; receives only relative audio signals , boundary-contact tactile flags, and the probed feature sample upon capture.
## Action space
- `move(dx, dy)`: where .
- `emit_sound(intensity)`: Evasive agents only, where .
- `tag()`: Seeker only. Attempts to capture any agent within radius .
- `probe_feature()`: Seeker only. Executed while holding a captive.
- `guess_id(agent_id)`: Seeker only. Submits identity guess for active captive.
## Playability notes
- **Single-player-vs-AI:** Highly playable as a sensory-filtered simulation. The human plays as the blindfolded seeker using spatial audio cues / radar pings and Bayesian inference over agent feature signatures.
- **Group games ():** Can be reduced to a 2-agent form (1 Seeker vs 1 Evader) by introducing simulated decoy noise emitters or static environmental obstacles to retain the audio localization challenge.
- **Physical-skill games:** Physical blindfolding and tactile touch are abstracted into an observation mask and a feature-sampling classifier with noise parameters.
- **What is lost in the adaptation:** The physical laughter, sudden tactile surprises, and communal warmth of childhood folk rhymes (*"Kanamachi bho-bho"*).
## Implementation hints
- Implement the Seeker as a Partially Observable Markov Decision Process (POMDP) agent maintaining a particle filter or Bayesian occupancy grid over evader positions.
- Branching factor: Seeker has 5 movement actions + tag/guess actions ( actions); evaders have actions each. Total joint action space , naturally decoupled in simultaneous-execution game engines.
- Suggested first milestone: Implement a grid with 1 blindfolded seeker and 3 evaders; verify that audio cues allow the seeker's particle filter to converge on evader coordinates.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Aankh Micholi (India). (n.d.). Games of Countries. https://gamesofcountries.com/games/in-india/aankh-micholi
@misc{gamesofcountries-in-india-aankh-micholi,
title = {Aankh Micholi (India)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/in-india/aankh-micholi}
}