Module games 

Game environment implementations

This module contains various game environments for reinforcement learning:

• CartPole: Classic cart-pole balancing task
• GridWorld: in-tree 4x4 FrozenLake-style sparse-reward navigation task (issue #182, i64 discrete action 0=Up/1=Right/2=Down/3=Left, 16-dim one-hot observation, absorbing goal/hole terminals + step-cap truncation)
• Snake: Snake game with configurable grid size
• SimpleBandit: Simple multi-armed bandit for testing
• Pong: Single-player Pong vs rule-based opponent
• ContinuousLqr: 1D LQR placeholder env that exercises the continuous (Vec<f32>) action surface added in issue #61
• PendulumSwingUp: in-tree Pendulum-v1 swing-up task; the canonical continuous-control SAC benchmark (issue #139, length-1 Vec<f32> torque action, 3-dim [cos θ, sin θ, θ̇] observation)
• MountainCarContinuous: in-tree MountainCarContinuous-v0 classic-control task; a sparse/deceptive-reward continuous-control benchmark (issue #166, length-1 Vec<f32> force action, 2-dim [position, velocity] observation, real terminal state at the goal)
• BucketBrigade (feature env-bucket-brigade): Slepian-Wolf MARL research env wrapping bucket_brigade_core with the versioned scenario registry from bucket-brigade PR #379
• MatchingPennies (feature training): 2-agent zero-sum smoke env implementing JointEnv for the multi-agent + PSRO trainers; canonical testbed for mixed-equilibrium learners (issue #107).
• NPlayerMatchingPennies (feature training): N-player “majority game” generalization. Each agent’s reward is +1 if its action matches the strict majority of other agents’ actions, -1 if against, 0 on tie (odd N only). Smoke env for N-player PSRO/NFSP testing (issue #119).

Re-exports

pub use cartpole::CartPole;

pub use continuous_lqr::ContinuousLqr;

pub use grid_world::GridWorld;

pub use matching_pennies::MatchingPennies;

pub use mountain_car_continuous::MountainCarContinuous;

pub use n_player_matching_pennies::NPlayerMatchingPennies;

pub use pendulum::PendulumSwingUp;

pub use pong::Pong;

pub use simple_bandit::SimpleBandit;

pub use snake::SnakeEnv;

Modules

cartpole

CartPole-v1 environment

continuous_lqr

Minimal continuous-action environment for trait-wiring proof.

grid_world

GridWorld sparse-reward discrete navigation environment.

matching_pennies

Matching Pennies — 2-agent zero-sum smoke env.

mountain_car_continuous

MountainCarContinuous classic-control environment.

n_player_matching_pennies

N-player matching pennies (“majority game”) — smoke env.

pendulum

Pendulum swing-up continuous-control environment.

pong

Single-player Pong environment.

simple_bandit

Simple Contextual Bandit Environment

snake

Multi-player Snake environment