fixes

Author: thowell

README.md

@@ -48,8 +48,8 @@ function particle(x, u, w)
 end
 
 # model
-dyn = Dynamics(particle, num_state, num_action)
-model = [dyn for t = 1:T-1] 
+dynamics = Dynamics(particle, num_state, num_action)
+model = [dynamics for t = 1:T-1] 
 
 # initialization
 x1 = [0.0; 0.0] 
@@ -62,17 +62,17 @@ ot = (x, u, w) -> 0.1 * dot(x, x) + 0.1 * dot(u, u)
 oT = (x, u, w) -> 0.1 * dot(x, x)
 ct = Cost(ot, num_state, num_action)
 cT = Cost(oT, num_state, 0)
-obj = [[ct for t = 1:T-1]..., cT]
+objective = [[ct for t = 1:T-1]..., cT]
 
 # constraints
 goal(x, u, w) = x - xT
 
 cont = Constraint()
 conT = Constraint(goal, num_state, 0)
-cons = [[cont for t = 1:T-1]..., conT] 
+constraints = [[cont for t = 1:T-1]..., conT] 
 
 # problem
-prob = Solver(model, obj, cons)
+prob = Solver(model, objective, constraints)
 initialize_controls!(prob, ū)
 initialize_states!(prob, x̄)
 

examples/acrobot.jl

@@ -89,8 +89,8 @@ function midpoint_explicit(x, u, w)
 end
 
 # ## model
-dyn = Dynamics(midpoint_explicit, num_state, num_action, num_parameter)
-model = [dyn for t = 1:T-1] 
+dynamics = Dynamics(midpoint_explicit, num_state, num_action, num_parameter)
+model = [dynamics for t = 1:T-1] 
 
 # ## initialization
 x1 = [0.0; 0.0; 0.0; 0.0] 
@@ -104,17 +104,17 @@ ot = (x, u, w) -> 0.1 * dot(x[3:4], x[3:4]) + 0.1 * dot(u, u)
 oT = (x, u, w) -> 0.1 * dot(x[3:4], x[3:4])
 ct = Cost(ot, num_state, num_action, num_parameter)
 cT = Cost(oT, num_state, 0, num_parameter)
-obj = [[ct for t = 1:T-1]..., cT]
+objective = [[ct for t = 1:T-1]..., cT]
 
 # ## constraints
 goal(x, u, w) = x - xT
 
 cont = Constraint()
 conT = Constraint(goal, num_state, 0)
-cons = [[cont for t = 1:T-1]..., conT] 
+constraints = [[cont for t = 1:T-1]..., conT] 
 
 # ## problem
-prob = Solver(model, obj, cons)
+prob = Solver(model, objective, constraints)
 initialize_controls!(prob, ū)
 initialize_states!(prob, x̄)
 

examples/car.jl

@@ -26,8 +26,8 @@ function midpoint_explicit(x, u, w)
 end
 
 # ## model
-dyn = Dynamics(midpoint_explicit, num_state, num_action, num_parameter)
-model = [dyn for t = 1:T-1] 
+dynamics = Dynamics(midpoint_explicit, num_state, num_action, num_parameter)
+model = [dynamics for t = 1:T-1] 
 
 # ## initialization
 x1 = [0.0; 0.0; 0.0] 
@@ -43,7 +43,7 @@ ot = (x, u, w) -> 1.0 * dot(x - xT, x - xT) + 1.0e-2 * dot(u, u)
 oT = (x, u, w) -> 1000.0 * dot(x - xT, x - xT)
 ct = Cost(ot, num_state, num_action, num_parameter)
 cT = Cost(oT, num_state, 0, num_parameter)
-obj = [[ct for t = 1:T-1]..., cT]
+objective = [[ct for t = 1:T-1]..., cT]
 
 # ## constraints
 ul = -5.0 * ones(num_action) 
@@ -71,10 +71,10 @@ end
 
 cont = Constraint(stage_con, num_state, num_action, indices_inequality=collect(1:5))
 conT = Constraint(terminal_con, num_state, num_action, indices_inequality=collect(3 .+ (1:1)))
-cons = [[cont for t = 1:T-1]..., conT] 
+constraints = [[cont for t = 1:T-1]..., conT] 
 
 # ## problem
-prob = Solver(model, obj, cons)
+prob = Solver(model, objective, constraints)
 initialize_controls!(prob, ū) 
 initialize_states!(prob, x̄)
 

examples/particle.jl

@@ -21,8 +21,8 @@ function particle(x, u, w)
 end
 
 # ## model
-dyn = Dynamics(particle, num_state, num_action)
-model = [dyn for t = 1:T-1] 
+dynamics = Dynamics(particle, num_state, num_action)
+model = [dynamics for t = 1:T-1] 
 
 # ## initialization
 x1 = [0.0; 0.0] 
@@ -35,17 +35,17 @@ ot = (x, u, w) -> 0.1 * dot(x, x) + 0.1 * dot(u, u)
 oT = (x, u, w) -> 0.1 * dot(x, x)
 ct = Cost(ot, num_state, num_action)
 cT = Cost(oT, num_state, 0)
-obj = [[ct for t = 1:T-1]..., cT]
+objective = [[ct for t = 1:T-1]..., cT]
 
 # ## constraints
 goal(x, u, w) = x - xT
 
 cont = Constraint()
 conT = Constraint(goal, num_state, 0)
-cons = [[cont for t = 1:T-1]..., conT] 
+constraints = [[cont for t = 1:T-1]..., conT] 
 
 # ## problem
-prob = Solver(model, obj, cons)
+prob = Solver(model, objective, constraints)
 initialize_controls!(prob, ū)
 initialize_states!(prob, x̄)
 

src/augmented_lagrangian.jl

@@ -36,25 +36,25 @@ function augmented_lagrangian(model::Model{T}, costs::Objective{T}, constraints:
         constraint_jacobian_action_tmp)
 end
 
-function cost(obj::AugmentedLagrangianCosts, states, actions, parameters)
+function cost(objective::AugmentedLagrangianCosts, states, actions, parameters)
     # costs
-    J = cost(obj.costs, states, actions, parameters)
+    J = cost(objective.costs, states, actions, parameters)
 
     # constraints
-    c = obj.constraint_data.violations
-    ρ = obj.constraint_penalty
-    λ = obj.constraint_dual
-    a = obj.active_set
+    c = objective.constraint_data.violations
+    ρ = objective.constraint_penalty
+    λ = objective.constraint_dual
+    a = objective.active_set
 
     # horizon
     H = length(c)
 
-    constraint!(obj.constraint_data, states, actions, parameters)
-    active_set!(a, obj.constraint_data, λ)
+    constraint!(objective.constraint_data, states, actions, parameters)
+    active_set!(a, objective.constraint_data, λ)
 
     for t = 1:H
         J += λ[t]' * c[t]
-        num_constraint = obj.constraint_data.constraints[t].num_constraint 
+        num_constraint = objective.constraint_data.constraints[t].num_constraint 
         for i = 1:num_constraint 
             if a[t][i] == 1
                 J += 0.5 * ρ[t][i] * c[t][i]^2.0
@@ -84,29 +84,29 @@ function active_set!(a, data::ConstraintsData, λ)
     end
 end
 
-function augmented_lagrangian_update!(obj::AugmentedLagrangianCosts;
+function augmented_lagrangian_update!(objective::AugmentedLagrangianCosts;
         scaling_penalty=10.0, 
         max_penalty=1.0e12)
 
     # constraints
-    c = obj.constraint_data.violations
-    cons = obj.constraint_data.constraints
-    ρ = obj.constraint_penalty
-    λ = obj.constraint_dual
+    c = objective.constraint_data.violations
+    constraints = objective.constraint_data.constraints
+    ρ = objective.constraint_penalty
+    λ = objective.constraint_dual
 
     # horizon
     H = length(c)
 
     for t = 1:H
-        num_constraint = cons[t].num_constraint 
+        num_constraint = constraints[t].num_constraint 
         for i = 1:num_constraint 
             λ[t][i] += ρ[t][i] * c[t][i]
-            if i in cons[t].indices_inequality
+            if i in constraints[t].indices_inequality
                 λ[t][i] = max(0.0, λ[t][i])
             end
             ρ[t][i] = min(scaling_penalty * ρ[t][i], max_penalty)
         end
     end
 end
 
-Base.length(obj::AugmentedLagrangianCosts) = length(obj.costs)
+Base.length(objective::AugmentedLagrangianCosts) = length(objective.costs)

src/data/constraints.jl

@@ -8,12 +8,12 @@ struct ConstraintsData{T,C,CX,CU}
     jacobian_action::Vector{CU}
 end
 
-function constraint_data(model::Model, cons::Constraints) 
-    H = length(cons)
-    c = [zeros(cons[t].num_constraint) for t = 1:H]
-    cx = [zeros(cons[t].num_constraint, t < H ? model[t].num_state : model[H-1].num_next_state) for t = 1:H]
-    cu = [zeros(cons[t].num_constraint, model[t].num_action) for t = 1:H-1]
-    ConstraintsData(cons, c, cx, cu)
+function constraint_data(model::Model, constraints::Constraints) 
+    H = length(constraints)
+    c = [zeros(constraints[t].num_constraint) for t = 1:H]
+    cx = [zeros(constraints[t].num_constraint, t < H ? model[t].num_state : model[H-1].num_next_state) for t = 1:H]
+    cu = [zeros(constraints[t].num_constraint, model[t].num_action) for t = 1:H-1]
+    ConstraintsData(constraints, c, cx, cu)
 end
 
 function constraint!(constraint_data::ConstraintsData, x, u, w)

src/data/solver.jl

@@ -61,7 +61,7 @@ end
 # TODO: fix iter
 function cache!(data::SolverData)
     iter = 1 #data.cache[:iter] 
-    # (iter > length(data[:obj])) && (@warn "solver data cache exceeded")
+    # (iter > length(data[:objective])) && (@warn "solver data cache exceeded")
     data.cache[:objective][iter] = data.objective[1]
     data.cache[:gradient][iter] = data.gradient
     data.cache[:max_violation][iter] = data.max_violation

src/gradients.jl

@@ -7,7 +7,7 @@ function gradients!(dynamics::Model, data::ProblemData;
     jacobian!(jx, ju, dynamics, x, u, w)
 end
 
-function gradients!(obj::Objective, data::ProblemData; 
+function gradients!(objective::Objective, data::ProblemData; 
         mode=:nominal)
     x, u, w = trajectories(data, 
         mode=mode)
@@ -16,11 +16,11 @@ function gradients!(obj::Objective, data::ProblemData;
     gxx = data.objective.hessian_state_state
     guu = data.objective.hessian_action_action
     gux = data.objective.hessian_action_state
-    cost_gradient!(gx, gu, obj, x, u, w)
-    cost_hessian!(gxx, guu, gux, obj, x, u, w) 
+    cost_gradient!(gx, gu, objective, x, u, w)
+    cost_hessian!(gxx, guu, gux, objective, x, u, w) 
 end
 
-function gradients!(obj::AugmentedLagrangianCosts, data::ProblemData; 
+function gradients!(objective::AugmentedLagrangianCosts, data::ProblemData; 
     mode=:nominal)
     # objective 
     gx = data.objective.gradient_state
@@ -30,29 +30,29 @@ function gradients!(obj::AugmentedLagrangianCosts, data::ProblemData;
     gux = data.objective.hessian_action_state
 
     # constraints
-    cons = obj.constraint_data.constraints
-    c = obj.constraint_data.violations
-    cx = obj.constraint_data.jacobian_state
-    cu = obj.constraint_data.jacobian_action
-    ρ = obj.constraint_penalty
-    λ = obj.constraint_dual
-    a = obj.active_set
-    Iρ = obj.constraint_penalty_matrix
-    c_tmp = obj.constraint_tmp 
-    cx_tmp = obj.constraint_jacobian_state_tmp 
-    cu_tmp = obj.constraint_jacobian_action_tmp
+    constraints = objective.constraint_data.constraints
+    c = objective.constraint_data.violations
+    cx = objective.constraint_data.jacobian_state
+    cu = objective.constraint_data.jacobian_action
+    ρ = objective.constraint_penalty
+    λ = objective.constraint_dual
+    a = objective.active_set
+    Iρ = objective.constraint_penalty_matrix
+    c_tmp = objective.constraint_tmp 
+    cx_tmp = objective.constraint_jacobian_state_tmp 
+    cu_tmp = objective.constraint_jacobian_action_tmp
 
     # horizon
-    H = length(obj)
+    H = length(objective)
 
     # derivatives
-    gradients!(obj.costs, data, 
+    gradients!(objective.costs, data, 
         mode=mode)
-    gradients!(obj.constraint_data, data, 
+    gradients!(objective.constraint_data, data, 
         mode=mode)
 
     for t = 1:H
-        num_constraint = cons[t].num_constraint
+        num_constraint = constraints[t].num_constraint
         for i = 1:num_constraint 
             Iρ[t][i, i] = ρ[t][i] * a[t][i]
         end

src/solver.jl

@@ -8,15 +8,15 @@ mutable struct Solver{T,N,M,NN,MM,MN,NNN,MNN,X,U,D,O,FX,FU,FW,OX,OU,OXX,OUU,OUX}
     options::Options{T}
 end
 
-function Solver(dynamics::Vector{Dynamics{T}}, obj::Objective{T}; 
+function Solver(dynamics::Vector{Dynamics{T}}, objective::Objective{T}; 
     parameters=[[zeros(d.num_parameter) for d in dynamics]..., zeros(0)],
     options=Options{T}()) where T
 
 	# allocate policy data
     policy = policy_data(dynamics)
 
     # allocate problem data
-    problem = problem_data(dynamics, obj, 
+    problem = problem_data(dynamics, objective, 
         parameters=parameters)
 
     # allocate solver data

test/acrobot.jl

@@ -80,8 +80,8 @@
     end
 
     # ## model
-    dyn = Dynamics(midpoint_explicit, num_state, num_action, num_parameter=num_parameter)
-    model = [dyn for t = 1:T-1] 
+    dynamics = Dynamics(midpoint_explicit, num_state, num_action, num_parameter=num_parameter)
+    model = [dynamics for t = 1:T-1] 
 
     # ## initialization
     x1 = [0.0; 0.0; 0.0; 0.0] 
@@ -95,14 +95,14 @@
     oT = (x, u, w) -> 0.1 * dot(x[3:4], x[3:4])
     ct = Cost(ot, num_state, num_action, num_parameter=num_parameter)
     cT = Cost(oT, num_state, 0, num_parameter=num_parameter)
-    obj = [[ct for t = 1:T-1]..., cT]
+    objective = [[ct for t = 1:T-1]..., cT]
 
     # ## constraints
     goal(x, u, w) = x - xT
 
     cont = Constraint()
     conT = Constraint(goal, num_state, 0)
-    cons = [[cont for t = 1:T-1]..., conT] 
+    constraints = [[cont for t = 1:T-1]..., conT] 
 
     # ## problem
     options = Options(verbose=false,
@@ -115,7 +115,7 @@
         initial_constraint_penalty=1.0,
         scaling_penalty=10.0)
 
-    s = Solver(model, obj, cons, options=options)
+    s = Solver(model, objective, constraints, options=options)
 
     initialize_controls!(s, ū) 
     initialize_states!(s, x̄)

test/benchmark.jl

@@ -29,13 +29,13 @@ rollout!(prob.policy, prob.problem, step_size=prob.solver_data.step_size[1])
 @benchmark rollout!($prob.policy, $prob.problem, step_size=$prob.solver_data.step_size[1])
 @code_warntype rollout!(prob.policy, prob.problem, step_size=prob.solver_data.step_size[1])
 
-augmented_lagrangian_update!(prob.problem.obj)
-@benchmark augmented_lagrangian_update!($prob.problem.obj)
-@code_warntype augmented_lagrangian_update!(prob.problem.obj)
+augmented_lagrangian_update!(prob.problem.objective)
+@benchmark augmented_lagrangian_update!($prob.problem.objective)
+@code_warntype augmented_lagrangian_update!(prob.problem.objective)
 
-constraint_violation(prob.problem.obj.constraint_data)
-@benchmark constraint_violation($prob.problem.obj.constraint_data)
-@code_warntype constraint_violation(prob.problem.obj.constraint_data)
+constraint_violation(prob.problem.objective.constraint_data)
+@benchmark constraint_violation($prob.problem.objective.constraint_data)
+@code_warntype constraint_violation(prob.problem.objective.constraint_data)
 
 initialize_controls!(prob, ū)
 @benchmark initialize_controls!($prob, $ū)
@@ -66,7 +66,7 @@ using BenchmarkTools
 @benchmark $A .= $(Diagonal(b .* a))
 
 t = 1
-a = prob.problem.obj.constraint_penalty[t]
-b = prob.problem.obj.a[t]
-c = prob.problem.obj.constraint_penalty_matrix[t]
+a = prob.problem.objective.constraint_penalty[t]
+b = prob.problem.objective.a[t]
+c = prob.problem.objective.constraint_penalty_matrix[t]
 @benchmark $c .= $(Diagonal(a .* b))