Skip to main content
Formal Specification

Hit the Pot — Formal Specification

Germany · guessing-hidden · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: discretizedGap-fills: 2
# Topfschlagen — AI-optimized specification ## General idea A spatial hidden-target navigation game under asymmetric information on a 2D bounded grid. A blindfolded Seeker agent has no visual perception of the target coordinate or grid boundaries, receiving only noisy 1D scalar acoustic distance-feedback tokens (*"Cold"*, *"Warm"*, *"Hot"*) and physical spoon tap collision events. The objective is to converge on the hidden pot coordinate in minimal temporal steps. ## Formal model - **Abstraction taken:** The continuous room is discretized into a bounded 2D grid [−W,W]×[−H,H] (standard: 11×11). The pot is placed at a hidden coordinate 𝐩pot=(xpot,ypot). The Seeker agent occupies state (x,y,θheading). The spoon tap spans a reach sector of radius Rtap=1.0 grid units in direction θheading. Distance feedback maps continuous Euclidean distance d=‖𝐩seeker−𝐩pot‖ into discrete acoustic tokens. - **Player/agent count:** 1 Seeker agent vs 1 Oracle/Environment (or N cooperative Guide agents). - **Turn model:** Strictly sequential search and tap action steps. ## Formal rules R1. **Grid & Initialization:** Grid dimensions [−5,5]2. Seeker starts at origin (0,0) with a random orientation θ0∈[0,2π). The pot position 𝐩pot is sampled uniformly at random from grid points with ‖𝐩pot‖≥2.0, hidden from the Seeker. Step counter t←0. R2. **Action Submission:** On each turn t, the Seeker submits one action from the legal action set: - `crawl(step_type)`: parameter `step_type` ∈{forward,turn_left,turn_right,reverse}. - `tap_spoon(sweep_angle)`: parameter `sweep_angle` ∈[−45.0∘,+45.0∘]. R3. **Crawl Kinematics & Boundary Clamp:** - `forward` updates position: (x′,y′)=(x+cos⁡θ,y+sin⁡θ). - Position is clamped to grid boundaries [−5,5]2. If a wall is struck, a tactile bump token `"Wall"` is returned. R4. **Acoustic Feedback Generation:** After any `crawl` action, the environment emits an acoustic distance token Token(d) where d=‖𝐩seeker′−𝐩pot‖: - If d>5.0: Token = `"Eiskalt"` - If 3.0<d≤5.0: Token = `"Kalt"` - If 1.5<d≤3.0: Token = `"Warm"` - If 0.5<d≤1.5: Token = `"Heiss"` - If d≤0.5: Token = `"Kochend Heiss"` R5. **Spoon Tap Collision Test:** When `tap_spoon(sweep_angle)` is submitted: - Tap test coordinate is 𝐩tap=𝐩seeker+Rtap⋅(cos⁡(θ+sweep),sin⁡(θ+sweep)). - If ‖𝐩tap−𝐩pot‖≤Rpot (where Rpot=0.5), a **Target Hit** event occurs (`CLANG!`). R6. **Victory Resolution:** A Target Hit immediately terminates the round with victory for the Seeker, recording total steps t. R7. [gap-fill] Search Step Limit / Stalemate Cap:** If the pot is not struck within Tmax=100 action steps, the round terminates with a timeout penalty. R8. [gap-fill] Multi-Player Tournament Metric:** In a multi-agent party tournament, agents are ranked in ascending order of steps taken to strike the pot (twin). ## Edge cases - Tie at the end condition: Tied step counts are broken by shortest cumulative path length traveled; if identical, joint victory is awarded. - A player with no legal move: Not applicable; rotation and tap actions are always legally available even if jammed against a boundary. - Repetition / cycles: Bounded strictly by hard step limit Tmax=100 in R7. - Simultaneous claims on the same resource or position: Single seeker per search trial; multi-player party formats run sequential trials with independent pot placements. - Player elimination mid-round; turn order repair: An agent disconnection terminates the active trial with a timeout score. - Resource exhaustion (pieces, tokens, space, time): Step counter strictly increments toward Tmax. - Termination guarantee: why the game provably ends: Guaranteed because step counter t strictly increments by 1 per action, reaching terminal state at t=Tmax if not resolved earlier by a Target Hit. ## State representation - **Persistent state:** Grid dimensions [−5,5]2, hidden pot coordinate 𝐩pot, step counter t, match score log. - **Per-player state:** Seeker position (x,y), heading angle θ, historical acoustic token log. - **Per-turn ephemeral state:** Submitted action, wall contact flag, emitted acoustic token, spoon hit test boolean. - **Observable information:** Imperfect information; the true pot position 𝐩pot is hidden from the Seeker. The Seeker observes self-orientation θ, tactile boundary contacts, previous actions, and the sequence of emitted scalar acoustic tokens. ## Action space - `crawl(step_type)`: spatial displacement/rotation. Parameter ∈{forward,turn_left,turn_right,reverse}. - `tap_spoon(sweep_angle)`: tactile sweep search. Parameter ∈[−45.0∘,+45.0∘]. Precondition: Seeker is active. ## Playability notes - Single-player-vs-AI: Engine-ready; excellent testbed for POMDP solvers, particle filters, Bayesian spatial inference, and reinforcement learning under sparse sensory cues. - Group games (n>4): Sequential trial structure scales cleanly across any number of party competitors. - Physical-skill games: Tactile crawling and acoustic gradient listening are mapped into discrete POMDP observation spaces. - What is lost in the adaptation: The tactile sensation of the floor carpet against knees, physical disorientation from spinning, and joyful nursery cheering. ## Implementation hints Extremely simple 2D grid POMDP environment. Implement as a standard Gym / PettingZoo environment in Python. Milestone 1: verify that a simple Bayesian particle filter updating posterior P(𝐩pot|Tokens) successfully converges on the pot in <25 steps.

Download as JSON

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

Download JSON
Cite this page

Hit the Pot (Germany). (n.d.). Games of Countries. https://gamesofcountries.com/games/de-germany/topfschlagen

@misc{gamesofcountries-de-germany-topfschlagen,
  title = {Hit the Pot (Germany)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/de-germany/topfschlagen}
}