Chaturanga

Chaturanga is a chess engine written in C++ from scratch. It currently takes a classical approach — relying on handcrafted evaluation and tree search.

Its architecture is built around pluggable networks, making it straightforward to introduce ML-based components down the line. It follows a three-network architecture:

• Policy Network — generates legal and candidate moves from a given position
• Search Tree — explores the game tree using depth-limited minimax
• Value Network — evaluates board positions to determine which side is winning

Each component is a pluggable interface — swap in a different evaluation strategy, move generator, or search algorithm without changing the rest of the engine. The same architecture could support a neural network evaluator or a learned move-ordering policy with minimal changes.

Named after Chaturanga, the ancient Indian game from which modern chess evolved.

Example Game

Human plays white, engine (depth 4, Stockfish piece-valuation-eval) plays black:

CLI:

  a b c d e f g h        e2e4       a b c d e f g h        d2d4       a b c d e f g h
8 r n b q k b n r 8      ───►     8 r n b q k b n r 8      ───►     8 r n b q k b n r 8
7 p p p p p p p p 7     b7b6      7 p . p p p p p p 7     a7a6      7 . . p p p p p p 7
6 . . . . . . . . 6               6 . p . . . . . . 6               6 p p . . . . . . 6
5 . . . . . . . . 5               5 . . . . . . . . 5               5 . . . . . . . . 5
4 . . . . . . . . 4               4 . . . . P . . . 4               4 . . . P P . . . 4
3 . . . . . . . . 3               3 . . . . . . . . 3               3 . . . . . . . . 3
2 P P P P P P P P 2               2 P P P P . P P P 2               2 P P P . . P P P 2
1 R N B Q K B N R 1               1 R N B Q K B N R 1               1 R N B Q K B N R 1
  a b c d e f g h                   a b c d e f g h                   a b c d e f g h

ScidvsPC:

< GUI >

Build

make

Usage

./chaturanga [options]

Options

Flag	Description	Default
--color <w|b>	Engine plays as white or black	b
--depth <int>	Search depth	4
--eval <type>	Evaluation: simple, kasparov, fisher, stockfish	stockfish
--fen <string>	Starting position in FEN notation	Standard starting position
--game <uci|cli>	Play via UCI-compatible GUI or interactive CLI	cli
--help	Show help message

Example

# Play as white against the engine (black) at depth 5
./chaturanga --color b --depth 5

# Play from a custom position
./chaturanga --fen "r1bqkbnr/pppppppp/2n5/8/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 1 2"

Move Format

Moves are entered in long algebraic notation: <from><to>[promotion]

• e2e4 — pawn to e4
• g1f3 — knight to f3
• e7e8q — pawn promotes to queen

Architecture

src/
├── main.cpp              # Entry point, CLI argument parsing
├── position/             # Board representation, move encoding
├── movegen/              # Legal move generation, attack detection
├── eval/                 # Position evaluation
├── search/               # Minimax search
└── game/                 # Game loop, Boss thread - I/O handling

Components

• Board — bitboard-based position representation with Zobrist hashing support
• MoveGen — pseudo-legal move generation with legality filtering
• Eval — material-based evaluation with multiple piece valuation systems
• Search — minimax depth-first search
• Game — manages the play loop between user and engine

Roadmap

Classical Improvements

• Alpha-beta pruning
• Iterative deepening
• Transposition tables
• Piece-square tables in evaluation
• Quiescence search
• Time management
• Opening book

Infrastructure

• UCI protocol support (CuteChess / SCID integration)
• Perft testing for move generation correctness
• FEN import/export

ML Integration

• NNUE evaluation (efficiently updatable neural net, replacing handcrafted eval)
• Policy network for move ordering (prioritize promising moves in search)
• Value network (deep neural net position evaluation, trained on Stockfish-labeled data)
• Monte Carlo Tree Search (MCTS guided by policy + value networks)

License

MIT