2021-12-12 23:19:18 +00:00
|
|
|
module RL
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-15 03:45:15 +00:00
|
|
|
export run_rl
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-10 02:16:45 +00:00
|
|
|
using ReinforcementLearning
|
2021-12-12 17:27:56 +00:00
|
|
|
using Flux: InvDecay
|
2021-12-12 14:29:08 +00:00
|
|
|
using Intervals
|
2021-12-12 17:27:56 +00:00
|
|
|
using StaticArrays: SVector
|
2021-12-14 03:03:14 +00:00
|
|
|
using LoopVectorization: @turbo
|
2021-12-12 23:19:18 +00:00
|
|
|
using Random: Random
|
2021-12-13 01:24:34 +00:00
|
|
|
using ProgressMeter: @showprogress
|
2021-12-12 23:19:18 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
using ..ReCo: ReCo, Particle, angle2, center_of_mass
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
const INITIAL_REWARD = 0.0
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
mutable struct Env <: AbstractEnv
|
|
|
|
n_actions::Int64
|
|
|
|
action_space::Vector{SVector{2,Float64}}
|
2021-12-14 03:03:14 +00:00
|
|
|
action_ind_space::Vector{Int64}
|
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
distance_state_space::Vector{Interval}
|
|
|
|
angle_state_space::Vector{Interval}
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
n_states::Int64
|
2021-12-20 23:31:44 +00:00
|
|
|
state_space::Vector{SVector{2,Interval}}
|
|
|
|
state_ind_space::Vector{Int64}
|
|
|
|
state_ind::Int64
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-10 02:16:45 +00:00
|
|
|
reward::Float64
|
2021-12-20 23:31:44 +00:00
|
|
|
terminated::Bool
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
center_of_mass::SVector{2,Float64}
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function Env(
|
2021-12-12 14:29:08 +00:00
|
|
|
min_distance::Float64,
|
|
|
|
max_distance::Float64;
|
2021-12-28 16:15:00 +00:00
|
|
|
n_v_actions::Int64=3,
|
|
|
|
n_ω_actions::Int64=5,
|
|
|
|
max_v::Float64=60.0,
|
2021-12-16 12:50:38 +00:00
|
|
|
max_ω::Float64=π / 2,
|
2021-12-28 16:15:00 +00:00
|
|
|
n_distance_states::Int64=3,
|
|
|
|
n_angle_states::Int64=4,
|
2021-12-10 02:16:45 +00:00
|
|
|
)
|
2021-12-12 14:29:08 +00:00
|
|
|
@assert min_distance > 0.0
|
|
|
|
@assert max_distance > min_distance
|
2021-12-10 02:16:45 +00:00
|
|
|
@assert n_v_actions > 1
|
2021-12-12 23:19:18 +00:00
|
|
|
@assert n_ω_actions > 1
|
2021-12-10 02:16:45 +00:00
|
|
|
@assert max_v > 0
|
|
|
|
@assert max_ω > 0
|
|
|
|
|
|
|
|
v_action_space = 0.0:(max_v / (n_v_actions - 1)):max_v
|
|
|
|
ω_action_space = (-max_ω):(2 * max_ω / (n_ω_actions - 1)):max_ω
|
|
|
|
|
|
|
|
n_actions = n_v_actions * n_ω_actions
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
action_space = Vector{SVector{2,Float64}}(undef, n_actions)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
|
|
|
ind = 1
|
|
|
|
for v in v_action_space
|
|
|
|
for ω in ω_action_space
|
2021-12-20 23:31:44 +00:00
|
|
|
action_space[ind] = SVector(v, ω)
|
2021-12-10 02:16:45 +00:00
|
|
|
ind += 1
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
action_ind_space = collect(1:n_actions)
|
2021-12-13 01:24:34 +00:00
|
|
|
|
2021-12-12 14:29:08 +00:00
|
|
|
distance_range =
|
|
|
|
min_distance:((max_distance - min_distance) / n_distance_states):max_distance
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
distance_state_space = Vector{Interval}(undef, n_distance_states)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
@simd for i in 1:n_distance_states
|
2021-12-12 14:29:08 +00:00
|
|
|
if i == 1
|
|
|
|
bound = Closed
|
|
|
|
else
|
|
|
|
bound = Open
|
|
|
|
end
|
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
distance_state_space[i] = Interval{Float64,bound,Closed}(
|
2021-12-12 14:29:08 +00:00
|
|
|
distance_range[i], distance_range[i + 1]
|
|
|
|
)
|
|
|
|
end
|
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
angle_range = (-π):(2 * π / n_angle_states):π
|
|
|
|
|
|
|
|
angle_state_space = Vector{Interval}(undef, n_angle_states)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
@simd for i in 1:n_angle_states
|
|
|
|
if i == 1
|
|
|
|
bound = Closed
|
|
|
|
else
|
|
|
|
bound = Open
|
|
|
|
end
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
angle_state_space[i] = Interval{Float64,bound,Closed}(
|
|
|
|
angle_range[i], angle_range[i + 1]
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
|
|
|
end
|
|
|
|
|
2021-12-16 13:54:52 +00:00
|
|
|
n_states = n_distance_states * n_angle_states + 1
|
2021-12-12 23:19:18 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
state_space = Vector{SVector{2,Interval}}(undef, n_states - 1)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
|
|
|
ind = 1
|
2021-12-12 14:29:08 +00:00
|
|
|
for distance_state in distance_state_space
|
2021-12-16 13:54:52 +00:00
|
|
|
for angle_state in angle_state_space
|
2021-12-20 23:31:44 +00:00
|
|
|
state_space[ind] = SVector(distance_state, angle_state)
|
2021-12-10 02:16:45 +00:00
|
|
|
ind += 1
|
|
|
|
end
|
|
|
|
end
|
2021-12-20 23:31:44 +00:00
|
|
|
# Last state is SVector(nothing, nothing)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
state_ind_space = collect(1:n_states)
|
2021-12-13 01:24:34 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# initial_state = SVector(nothing, nothing)
|
|
|
|
initial_state_ind = n_states
|
|
|
|
|
2021-12-12 14:29:08 +00:00
|
|
|
return new(
|
2021-12-20 23:31:44 +00:00
|
|
|
n_actions,
|
2021-12-13 01:24:34 +00:00
|
|
|
action_space,
|
2021-12-14 03:03:14 +00:00
|
|
|
action_ind_space,
|
2021-12-13 01:24:34 +00:00
|
|
|
distance_state_space,
|
2021-12-16 13:54:52 +00:00
|
|
|
angle_state_space,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_states,
|
2021-12-13 01:24:34 +00:00
|
|
|
state_space,
|
2021-12-14 03:03:14 +00:00
|
|
|
state_ind_space,
|
2021-12-20 23:31:44 +00:00
|
|
|
initial_state_ind,
|
2021-12-14 03:03:14 +00:00
|
|
|
INITIAL_REWARD,
|
2021-12-20 23:31:44 +00:00
|
|
|
false,
|
2021-12-28 16:15:00 +00:00
|
|
|
SVector(0.0, 0.0),
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
2021-12-10 02:16:45 +00:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function reset!(env::Env)
|
|
|
|
env.state_ind = env.n_states
|
|
|
|
env.reward = INITIAL_REWARD
|
|
|
|
env.terminated = false
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
RLBase.state_space(env::Env) = env.state_ind_space
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
RLBase.state(env::Env) = env.state_ind
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
RLBase.action_space(env::Env) = env.action_ind_space
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
RLBase.reward(env::Env) = env.reward
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
RLBase.is_terminated(env::Env) = env.terminated
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
struct Params{H<:AbstractHook}
|
|
|
|
env::Env
|
|
|
|
agent::Agent
|
|
|
|
hook::H
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
old_states_ind::Vector{Int64}
|
|
|
|
states_ind::Vector{Int64}
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
actions::Vector{SVector{2,Float64}}
|
|
|
|
actions_ind::Vector{Int64}
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_steps_before_actions_update::Int64
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
goal_shape_ratio::Float64
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles::Int64
|
2021-12-14 03:03:14 +00:00
|
|
|
min_sq_distances::Vector{Float64}
|
|
|
|
vecs_r⃗₁₂_to_min_distance_particle::Vector{SVector{2,Float64}}
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
half_box_len::Float64
|
|
|
|
max_elliptic_distance::Float64
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function Params(
|
|
|
|
env::Env,
|
|
|
|
agent::Agent,
|
|
|
|
hook::H,
|
2021-12-13 01:24:34 +00:00
|
|
|
n_steps_before_actions_update::Int64,
|
|
|
|
goal_shape_ratio::Float64,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles::Int64,
|
2021-12-28 16:15:00 +00:00
|
|
|
half_box_len::Float64,
|
2021-12-12 17:27:56 +00:00
|
|
|
) where {H<:AbstractHook}
|
2021-12-28 16:15:00 +00:00
|
|
|
max_elliptic_distance = sqrt(2) * half_box_len
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_states = env.n_states
|
|
|
|
|
|
|
|
return new{H}(
|
|
|
|
env,
|
|
|
|
agent,
|
|
|
|
hook,
|
|
|
|
fill(0, n_particles),
|
|
|
|
fill(n_states, n_particles),
|
|
|
|
fill(SVector(0.0, 0.0), n_particles),
|
|
|
|
fill(0, n_particles),
|
2021-12-12 14:29:08 +00:00
|
|
|
n_steps_before_actions_update,
|
2021-12-13 01:24:34 +00:00
|
|
|
goal_shape_ratio,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles,
|
|
|
|
fill(Inf64, n_particles),
|
|
|
|
fill(SVector(0.0, 0.0), n_particles),
|
2021-12-28 16:15:00 +00:00
|
|
|
half_box_len,
|
|
|
|
max_elliptic_distance,
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function pre_integration_hook(rl_params::Params)
|
|
|
|
@turbo for i in 1:(rl_params.n_particles)
|
2021-12-14 03:03:14 +00:00
|
|
|
rl_params.min_sq_distances[i] = Inf64
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-12 17:27:56 +00:00
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function state_update_helper_hook(
|
|
|
|
rl_params::Params, id1::Int64, id2::Int64, r⃗₁₂::SVector{2,Float64}, distance²::Float64
|
2021-12-12 17:27:56 +00:00
|
|
|
)
|
2021-12-14 03:03:14 +00:00
|
|
|
if rl_params.min_sq_distances[id1] > distance²
|
|
|
|
rl_params.min_sq_distances[id1] = distance²
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
rl_params.vecs_r⃗₁₂_to_min_distance_particle[id1] = r⃗₁₂
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
if rl_params.min_sq_distances[id2] > distance²
|
|
|
|
rl_params.min_sq_distances[id2] = distance²
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
rl_params.vecs_r⃗₁₂_to_min_distance_particle[id2] = -r⃗₁₂
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function get_state_ind(state::S, state_space::Vector{S}) where {S<:SVector{2,Interval}}
|
|
|
|
return findfirst(x -> x == state, state_space)
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
function state_update_hook(rl_params::Params, particles::Vector{Particle})
|
|
|
|
@turbo for id in 1:(rl_params.n_particles)
|
|
|
|
rl_params.old_states_ind[id] = rl_params.states_ind[id]
|
2021-12-20 23:31:44 +00:00
|
|
|
end
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
env = rl_params.env
|
2021-12-13 01:24:34 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_states = env.n_states
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
env_angle_state = env.angle_state_space[1]
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
state_space = env.state_space
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
for id in 1:(rl_params.n_particles)
|
2021-12-16 13:54:52 +00:00
|
|
|
env_distance_state::Union{Interval,Nothing} = nothing
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
min_sq_distance = rl_params.min_sq_distances[id]
|
2021-12-14 03:03:14 +00:00
|
|
|
min_distance = sqrt(min_sq_distance)
|
|
|
|
|
|
|
|
if !isinf(min_sq_distance)
|
2021-12-20 23:31:44 +00:00
|
|
|
for distance_state in env.distance_state_space
|
2021-12-16 13:54:52 +00:00
|
|
|
if min_distance in distance_state
|
2021-12-14 03:03:14 +00:00
|
|
|
env_distance_state = distance_state
|
|
|
|
break
|
|
|
|
end
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# (nothing, nothing)
|
|
|
|
state_ind = n_states
|
|
|
|
|
|
|
|
if !isnothing(env_distance_state)
|
2021-12-28 16:15:00 +00:00
|
|
|
r⃗₁₂ = rl_params.vecs_r⃗₁₂_to_min_distance_particle[id]
|
|
|
|
si, co = sincos(particles[id].φ)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
|
|
|
#=
|
|
|
|
Angle between two vectors
|
|
|
|
e = (co, si)
|
|
|
|
angle = acos(dot(r⃗₁₂, e) / (norm(r⃗₁₂) * norm(e)))
|
|
|
|
norm(r⃗₁₂) == min_distance
|
|
|
|
norm(e) == 1
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
min_distance is not infinite, because otherwise
|
2021-12-16 13:54:52 +00:00
|
|
|
env_angle_state would be nothing and this else block will not be called
|
2021-12-12 17:27:56 +00:00
|
|
|
=#
|
2021-12-16 13:54:52 +00:00
|
|
|
angle = angle2(SVector(co, si), r⃗₁₂)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
for angle_state in env.angle_state_space
|
2021-12-16 13:54:52 +00:00
|
|
|
if angle in angle_state
|
|
|
|
env_angle_state = angle_state
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
state = SVector{2,Interval}(env_distance_state, env_angle_state)
|
|
|
|
state_ind = get_state_ind(state, state_space)
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
rl_params.states_ind[id] = state_ind
|
2021-12-20 23:31:44 +00:00
|
|
|
end
|
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
env.center_of_mass = center_of_mass(particles, rl_params.half_box_len)
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
|
|
|
function get_env_agent_hook(rl_params::Params)
|
|
|
|
return (rl_params.env, rl_params.agent, rl_params.hook)
|
|
|
|
end
|
|
|
|
|
|
|
|
function update_table_and_actions_hook(
|
|
|
|
rl_params::Params, particle::Particle, first_integration_step::Bool
|
|
|
|
)
|
|
|
|
env, agent, hook = get_env_agent_hook(rl_params)
|
|
|
|
|
|
|
|
id = particle.id
|
|
|
|
|
|
|
|
if !first_integration_step
|
|
|
|
# Old state
|
|
|
|
env.state_ind = rl_params.old_states_ind[id]
|
|
|
|
|
|
|
|
action_ind = rl_params.actions_ind[id]
|
|
|
|
|
|
|
|
# Pre act
|
|
|
|
agent(PRE_ACT_STAGE, env, action_ind)
|
|
|
|
hook(PRE_ACT_STAGE, agent, env, action_ind)
|
|
|
|
|
|
|
|
# Update to current state
|
|
|
|
env.state_ind = rl_params.states_ind[id]
|
|
|
|
|
|
|
|
# Update reward
|
2021-12-28 16:15:00 +00:00
|
|
|
vec_to_center_of_mass = ReCo.minimum_image(
|
|
|
|
particle.c - env.center_of_mass, rl_params.half_box_len
|
|
|
|
)
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
env.reward =
|
|
|
|
-(vec_to_center_of_mass[1]^2 + vec_to_center_of_mass[2]^2) /
|
|
|
|
rl_params.max_elliptic_distance / rl_params.n_particles
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Post act
|
2021-12-12 17:27:56 +00:00
|
|
|
agent(POST_ACT_STAGE, env)
|
2021-12-14 03:03:14 +00:00
|
|
|
hook(POST_ACT_STAGE, agent, env)
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# Update action
|
|
|
|
action_ind = agent(env)
|
|
|
|
action = env.action_space[action_ind]
|
|
|
|
|
|
|
|
rl_params.actions[id] = action
|
|
|
|
rl_params.actions_ind[id] = action_ind
|
2021-12-28 16:15:00 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
|
|
|
act_hook(::Nothing, args...) = nothing
|
|
|
|
|
|
|
|
function act_hook(
|
|
|
|
rl_params::Params, particle::Particle, δt::Float64, si::Float64, co::Float64
|
|
|
|
)
|
|
|
|
# Apply action
|
|
|
|
action = rl_params.actions[particle.id]
|
|
|
|
|
|
|
|
vδt = action[1] * δt
|
|
|
|
particle.tmp_c += SVector(vδt * co, vδt * si)
|
|
|
|
particle.φ += action[2] * δt
|
|
|
|
|
2021-12-12 17:27:56 +00:00
|
|
|
return nothing
|
|
|
|
end
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
function gen_agent(n_states::Int64, n_actions::Int64, ϵ::Float64)
|
2021-12-20 23:31:44 +00:00
|
|
|
policy = QBasedPolicy(;
|
|
|
|
learner=MonteCarloLearner(;
|
|
|
|
approximator=TabularQApproximator(;
|
|
|
|
n_state=n_states, n_action=n_actions, opt=InvDecay(1.0)
|
|
|
|
),
|
|
|
|
),
|
2021-12-28 16:15:00 +00:00
|
|
|
explorer=EpsilonGreedyExplorer(ϵ),
|
2021-12-20 23:31:44 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
return Agent(; policy=policy, trajectory=VectorSARTTrajectory())
|
|
|
|
end
|
|
|
|
|
2021-12-15 03:45:15 +00:00
|
|
|
function run_rl(;
|
2021-12-13 01:24:34 +00:00
|
|
|
goal_shape_ratio::Float64,
|
2021-12-28 16:15:00 +00:00
|
|
|
n_episodes::Int64=200,
|
2021-12-14 03:03:14 +00:00
|
|
|
episode_duration::Float64=50.0,
|
2021-12-28 16:15:00 +00:00
|
|
|
update_actions_at::Float64=0.1,
|
2021-12-13 01:24:34 +00:00
|
|
|
n_particles::Int64=100,
|
2021-12-14 03:03:14 +00:00
|
|
|
seed::Int64=42,
|
2021-12-28 16:15:00 +00:00
|
|
|
ϵ::Float64=0.01,
|
|
|
|
parent_dir::String="",
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
2021-12-13 01:24:34 +00:00
|
|
|
@assert 0.0 <= goal_shape_ratio <= 1.0
|
2021-12-12 14:29:08 +00:00
|
|
|
@assert n_episodes > 0
|
|
|
|
@assert episode_duration > 0
|
2021-12-28 16:15:00 +00:00
|
|
|
@assert update_actions_at in 0.001:0.001:episode_duration
|
2021-12-12 14:29:08 +00:00
|
|
|
@assert n_particles > 0
|
2021-12-28 16:15:00 +00:00
|
|
|
@assert 0.0 < ϵ < 1.0
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Setup
|
|
|
|
Random.seed!(seed)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
sim_consts = ReCo.gen_sim_consts(
|
|
|
|
n_particles, 0.0; skin_to_interaction_r_ratio=1.8, packing_ratio=0.15
|
|
|
|
)
|
2021-12-12 23:19:18 +00:00
|
|
|
n_particles = sim_consts.n_particles
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
env = Env(sim_consts.particle_radius, sim_consts.skin_r)
|
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
agent = gen_agent(env.n_states, env.n_actions, ϵ)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-12 23:19:18 +00:00
|
|
|
n_steps_before_actions_update = round(Int64, update_actions_at / sim_consts.δt)
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
hook = TotalRewardPerEpisode()
|
|
|
|
|
|
|
|
rl_params = Params(
|
2021-12-28 16:15:00 +00:00
|
|
|
env,
|
|
|
|
agent,
|
|
|
|
hook,
|
|
|
|
n_steps_before_actions_update,
|
|
|
|
goal_shape_ratio,
|
|
|
|
n_particles,
|
|
|
|
sim_consts.half_box_len,
|
2021-12-12 17:27:56 +00:00
|
|
|
)
|
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
parent_dir = "RL" * parent_dir
|
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Pre experiment
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(PRE_EXPERIMENT_STAGE, agent, env)
|
|
|
|
agent(PRE_EXPERIMENT_STAGE, env)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-13 01:24:34 +00:00
|
|
|
@showprogress 0.6 for episode in 1:n_episodes
|
2021-12-28 16:15:00 +00:00
|
|
|
dir = ReCo.init_sim_with_sim_consts(sim_consts; parent_dir=parent_dir)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Reset
|
2021-12-20 23:31:44 +00:00
|
|
|
reset!(env)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Pre espisode
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(PRE_EPISODE_STAGE, agent, env)
|
|
|
|
agent(PRE_EPISODE_STAGE, env)
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
# Episode
|
2021-12-16 12:50:38 +00:00
|
|
|
ReCo.run_sim(
|
2021-12-12 17:27:56 +00:00
|
|
|
dir; duration=episode_duration, seed=rand(1:typemax(Int64)), rl_params=rl_params
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
env.terminated = true
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# Post episode
|
|
|
|
hook(POST_EPISODE_STAGE, agent, env)
|
|
|
|
agent(POST_EPISODE_STAGE, env)
|
2021-12-28 16:15:00 +00:00
|
|
|
|
|
|
|
display(hook.rewards)
|
2021-12-12 14:29:08 +00:00
|
|
|
end
|
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Post experiment
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(POST_EXPERIMENT_STAGE, agent, env)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-13 01:24:34 +00:00
|
|
|
return rl_params
|
2021-12-12 14:29:08 +00:00
|
|
|
end
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-12 14:29:08 +00:00
|
|
|
end # module
|