Crate rl_traits 

Core traits for reinforcement learning environments, policies, and agents.

rl-traits defines the shared vocabulary used across the ecosystem:

• [ember-rl]: algorithm implementations using Burn for neural networks
• [bevy-gym]: Bevy ECS plugin for visualising and parallelising environments

Design goals

• Type-safe by default: observation and action spaces are associated types, not runtime objects. The compiler catches mismatches.

• Correct Terminated vs Truncated distinction: algorithms that bootstrap value estimates (PPO, DQN, SAC) need this distinction. It is baked into EpisodeStatus from day one.

• Rendering-free: this crate has no concept of visualisation. That belongs in bevy-gym.

• Bevy-compatible: Send + Sync + 'static bounds on associated types mean any Environment implementation can be a Bevy Component, enabling free ECS parallelisation via Query::par_iter_mut().

• Minimal dependencies: only rand for RNG abstractions.

Quick start

use rl_traits::{Environment, StepResult, EpisodeStatus};
use rand::Rng;

struct MyEnv;

impl Environment for MyEnv {
    type Observation = f32;
    type Action = usize;
    type Info = ();

    fn step(&mut self, action: usize) -> StepResult<f32, ()> {
        StepResult::new(0.0, 1.0, EpisodeStatus::Continuing, ())
    }

    fn reset(&mut self, _seed: Option<u64>) -> (f32, ()) {
        (0.0, ())
    }

    fn sample_action(&self, rng: &mut impl Rng) -> usize {
        rng.gen_range(0..4)
    }
}

Re-exports

pub use agent::Agent;

pub use buffer::ReplayBuffer;

pub use environment::Environment;

pub use episode::EpisodeStatus;

pub use episode::StepResult;

pub use experience::Experience;

pub use multi_agent::AecEnvironment;

pub use multi_agent::ParallelEnvironment;

pub use policy::Policy;

pub use policy::StochasticPolicy;

pub use wrappers::TimeLimit;

pub use wrappers::Wrapper;

Modules

agent

buffer

environment

episode

experience

multi_agent

policy

wrappers