Towards ReinforcementLearning.jl integration

Author: rejuvyesh

Project.toml

@@ -19,6 +19,7 @@ Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 Polyhedra = "67491407-f73d-577b-9b50-8179a7c68029"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+ReinforcementLearningBase = "e575027e-6cd6-5018-9292-cdc6200d2b44"
 Rotations = "6038ab10-8711-5258-84ad-4b1120ba62dc"
 Scratch = "6c6a2e73-6563-6170-7368-637461726353"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"

environments/rlenv.jl

@@ -0,0 +1,38 @@
+using ReinforcementLearningBase: RLBase
+
+mutable struct DojoRLEnv <: RLBase.AbstractEnv
+    dojoenv
+    action_space
+    observation_space
+    state
+    reward
+    done::Bool
+    info::Dict
+end
+
+function DojoRLEnv(dojoenv::Environment)
+    action_space = convert(RLBase.Space, dojoenv.input_space)
+    observation_space = convert(RLBase.Space, dojoenv.observation_space)
+    state = reset(dojoenv)
+    return DojoRLEnv(dojoenv, action_space, observation_space, state, 0.0, false, Dict())
+end
+
+RLBase.action_space(env::DojoRLEnv) = env.action_space
+RLBase.state_space(env::DojoRLEnv) = env.observation_space
+RLBase.is_terminated(env::DojoRLEnv) = env.done
+
+RLBase.reset!(env::DojoRLEnv) = reset(env.dojoenv)
+
+RLBase.reward(env::DojoRLEnv) = error()
+RLBase.state(env::DojoRLEnv) = env.state
+
+Random.seed!(env::DojoRLEnv, seed) = Dojo.seed(env.dojoenv, seed)
+
+function (env::DojoRLEnv)(a)
+    s, r, d, i = step(env.dojoenv, a)
+    env.state = s
+    env.reward = r
+    env.done = d
+    env.info = i
+    return nothing
+end

src/Dojo.jl

@@ -146,6 +146,7 @@ include(joinpath("..", "environments", "environment.jl"))
 include(joinpath("..", "environments", "dynamics.jl"))
 include(joinpath("..", "environments", "utilities.jl"))
 include(joinpath("..", "environments", "include.jl"))
+include(joinpath("..", "environments", "rlenv.jl"))
 
 # Bodies
 export