Merge branch 'main' into compat_for_1_7

Author: janbruedigam

.github/workflows/CI.yml

@@ -13,7 +13,7 @@ jobs:
     strategy:
       matrix:
         version:
-          - '1.6'
+          - '1.7'
         os:
           - ubuntu-latest
 #           - macOS-latest
@@ -26,11 +26,6 @@ jobs:
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
-#       - name: Add registries
-#         run: |
-#           julia --color=yes -e 'using Pkg; pkg"registry add https://github.com/JuliaRegistries/General.git"'
-#           julia --color=yes -e 'using Pkg; pkg"registry add https://github.com/Lyceum/LyceumRegistry.git"'
-#         shell: bash
       - uses: julia-actions/julia-runtest@latest
       - uses: julia-actions/julia-uploadcodecov@latest
         env:

.github/workflows/TagBot.yml

@@ -0,0 +1,15 @@
+name: TagBot
+on:
+  issue_comment:
+    types:
+      - created
+  workflow_dispatch:
+jobs:
+  TagBot:
+    if: github.event_name == 'workflow_dispatch' || github.actor == 'JuliaTagBot'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: JuliaRegistries/TagBot@v1
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          ssh: ${{ secrets.DOCUMENTER_KEY }}

README.md

@@ -3,11 +3,12 @@
 [![](https://img.shields.io/badge/docs-dev-blue.svg)](https://dojo-sim.github.io/Dojo.jl/dev)
 
 # Dojo
-A differentiable physics simulator for robotics
+A differentiable simulator for robotics
 - arXiv preprint: https://arxiv.org/abs/2203.00806
 - Python interface: https://github.com/dojo-sim/dojopy
-- Site: https://sites.google.com/view/dojo-sim
-# Examples (RSS 2022)
+- site: https://sites.google.com/view/dojo-sim
+
+# Examples
 
 ## Simulation
 

environments/ant/methods/env.jl

@@ -76,7 +76,7 @@ function ant(;
     return env
 end
 
-function step(env::Environment{Ant}, x, u; 
+function Base.step(env::Environment{Ant}, x, u; 
     gradients=false,
     attitude_decompress=false)
 

environments/pendulum/methods/env.jl

@@ -99,7 +99,7 @@ function get_observation(env::Environment{Pendulum})
     end
 end
 
-function step(env::Environment{Pendulum}, x, u; 
+function Base.step(env::Environment{Pendulum}, x, u; 
     gradients=false,
     attitude_decompress=false)
     mechanism = env.mechanism

examples/Project.toml

@@ -1,12 +1,14 @@
 [deps]
 Dojo = "ac60b53e-8d92-4c83-b960-e78698fa1916"
+Folds = "41a02a25-b8f0-4f67-bc48-60067656b558"
 IterativeLQR = "605048dd-e178-462b-beb9-98a09398ef27"
 JLD2 = "033835bb-8acc-5ee8-8aae-3f567f8a3819"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 PGFPlots = "3b7a836e-365b-5785-a47d-02c71176b4aa"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 
 [compat]
+Folds = "0.2"
 IterativeLQR = "0.1.1"
 JLD2 = "0.4.21"
 Literate = "2.13.0"

examples/reinforcement_learning/algorithms/ars.jl

@@ -3,12 +3,13 @@
 ################################################################################
 using LinearAlgebra
 using Statistics
+using Folds
 
 import LinearAlgebra.normalize
-import GeometryBasics.update
+import Dojo.GeometryBasics.update
 
 # ARS options: hyper parameters
-@with_kw struct HyperParameters{T}
+Base.@kwdef struct HyperParameters{T}
     main_loop_size::Int = 100
     horizon::Int = 200
     step_size::T = 0.02
@@ -114,16 +115,19 @@ function rollout_policy(θ::Matrix, env::Environment, normalizer::Normalizer, hp
 end
 
 function train(env::Environment, policy::Policy{T}, normalizer::Normalizer{T},
-        hp::HyperParameters{T}; distributed=false) where T
+        hp::HyperParameters{T}; distributed=false, usefolds=false, foldsexec=Folds.ThreadedEx(;basesize=1)) where T
     println("Training linear policy with Augmented Random Search (ARS)\n ")
     if distributed
         envs = [deepcopy(env) for i = 1:(2 * hp.n_directions)]
         normalizers = [deepcopy(normalizer) for i = 1:(2 * hp.n_directions)]
         hps = [deepcopy(hp) for i = 1:(2 * hp.n_directions)]
         print("  $(nprocs()) processors")
+    elseif usefolds
+        envs = [deepcopy(env) for i = 1:(2*hp.n_directions)]
+        print("  $(Threads.nthreads()) threads with Folds")
     else
-        envs = [deepcopy(env) for i = 1:Threads.nthreads()]
-        print("  $(Threads.nthreads()) threads")
+        envs = [deepcopy(env) for i = 1:(Threads.nthreads())]
+        print(" $(Threads.nthreads()) ")
     end
 
     # pre-allocate for rewards
@@ -134,14 +138,20 @@ function train(env::Environment, policy::Policy{T}, normalizer::Normalizer{T},
         θs, δs = sample_policy(policy)
 
         # evaluate policies
-        if distributed
-            rewards .= pmap(rollout_policy, θs, envs, normalizers, hps)
-        else
-            Threads.@threads for k = 1:(2 * hp.n_directions)
-                rewards[k] = rollout_policy(θs[k], envs[Threads.threadid()], normalizer, hp)
+        roll_time = @elapsed begin
+            if distributed
+                rewards .= pmap(rollout_policy, θs, envs, normalizers, hps)
+            elseif usefolds
+                @assert length(envs) == size(θs, 1) "$(length(envs))"
+                Folds.map!(rewards, θs, envs, foldsexec) do θ, env
+                    rollout_policy(θ, env, normalizer, hp)
+                end
+            else
+                Threads.@threads for k = 1:(2 * hp.n_directions)
+                    rewards[k] = rollout_policy(θs[k], envs[Threads.threadid()], normalizer, hp)
+                end
             end
         end
-
         # reward evaluation
         r_max = [max(rewards[k], rewards[hp.n_directions + k]) for k = 1:hp.n_directions]
         σ_r = std(rewards)
@@ -152,7 +162,7 @@ function train(env::Environment, policy::Policy{T}, normalizer::Normalizer{T},
         update(policy, rollouts, σ_r)
 
         # finish, print:
-        println("episode $episode reward_evaluation $(mean(rewards))")
+        println("episode $episode reward_evaluation $(mean(rewards)). Took $(roll_time) seconds")
     end
 
     return nothing