Merge pull request #17 from RoboticExplorationLab/newaltro

Changes for new version of Altro

Author: bjack205

NEWS.md

@@ -1,3 +1,12 @@
+# New `v0.4`
+Completely new overhaul of the existing API. Now allows for both "inplace" and 
+"out-of-place" methods that work with StaticArrays. The method for differentiation 
+can be chosen at runtime. No methods are provided by default, but the `@autodiff` macro is 
+provided which implements forward automatic differentiation via [ForwardDiff.jl](https://github.com/JuliaDiff/ForwardDiff.jl) and finite 
+differencing via [FiniteDiff.jl](https://github.com/JuliaDiff/FiniteDiff.jl).
+
+Support for linear models is no longer provided. This functionality is likely to appear in 
+a downstream package such as [RobotZoo.jl](https://github.com/RoboticExplorationLab/RobotZoo.jl).
 # New in v0.3
 ## `integrate` method
 Allow for more flexibility for defining discrete dynamics by using 

Project.toml

@@ -1,7 +1,7 @@
 name = "RobotDynamics"
 uuid = "38ceca67-d8d3-44e8-9852-78a5596522e1"
 authors = ["Brian Jackson <[email protected]>"]
-version = "0.4.2"
+version = "0.4.3"
 
 [deps]
 FiniteDiff = "6a86dc24-6348-571c-b903-95158fe2bd41"

README.md

@@ -65,7 +65,7 @@ RobotDynamics.control_dim(::Cartpole) = 1
 
 # Create the model
 model = Cartpole()
-n,m = size(model)
+n,m = RD.dims(model)
 
 # Generate random state and control vector
 x,u = rand(model)

docs/Project.toml

@@ -1,5 +1,8 @@
 [deps]
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
+FiniteDiff = "6a86dc24-6348-571c-b903-95158fe2bd41"
+ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 
 [compat]
 Documenter = "0.24"

docs/make.jl

@@ -1,24 +1,45 @@
 using Documenter
 using RobotDynamics
+using FiniteDiff
+using ForwardDiff
+using StaticArrays
 
 makedocs(
     sitename = "RobotDynamics",
     format = Documenter.HTML(prettyurls = false),
     pages = [
         "Introduction" => "index.md",
-        "Getting Started" => [
-            "models.md",
-            "dynamics_evaluation.md"
+        "Dynamics Models" => [
+            "continuous.md",
+            "discrete.md",
         ],
-        "Documentation" => [
-            "discretization.md",
+        "Rotations" => [
+            # "liestate.md"
+            "rotationstate.md",
+        ],
+        "Important Types" => [
             "knotpoints.md",
-            "liemodel.md",
-            "rigidbody.md",
-            "linearmodel.md",
-            "finite_diff.md",
-            "dynamics_api.md"   
-        ]
+            "trajectories.md",
+        ],
+        "Internal API" => [
+            "functionbase.md",
+            "scalarfunction.md",
+            "autodiff.md",
+        ],
+
+        # "Getting Started" => [
+        #     "models.md",
+        #     "dynamics_evaluation.md"
+        # ],
+        # "Documentation" => [
+        #     "discretization.md",
+        #     "knotpoints.md",
+        #     "liemodel.md",
+        #     "rigidbody.md",
+        #     "linearmodel.md",
+        #     "finite_diff.md",
+        #     "dynamics_api.md"   
+        # ]
     ]
 )
 

docs/src/discretization.md renamed to docs/oldsrc/discretization.md

@@ -0,0 +1,41 @@
+# `KnotPoint` type
+```@meta
+CurrentModule = RobotDynamics
+```
+
+A common way of dealing with trajectories of forced dynamical systems, especially in optimization,
+is to represent a trajectory with a fixed number of "knot points", typically distributed
+evenly over time. Each point records the states, controls, time, and time step to the next
+point. It is often convenient to store all this information together, which is the purpose
+of the `AbstractKnotPoint` type. Additionally, it is almost always more efficient to index
+into a concatenated vector than it is to concatenate two smaller vectors, so the states
+and controls are stacked together in a single  `n+m`-dimensional vector.
+
+RobotDynamics.jl defines a couple different implementations of the `AbstractKnotPoint`
+interface, which can be useful depending on the application.
+
+## Types
+```@docs
+AbstractKnotPoint
+GeneralKnotPoint
+KnotPoint
+StaticKnotPoint
+```
+
+## Methods
+All `AbstractKnotPoint` types support the following methods:
+
+```@docs
+state
+control
+is_terminal
+get_z
+set_state!
+set_control!
+set_z!
+```
+
+## Mathematical Operations
+All `AbstractKnotPoint` types support addition between two knot points, addition of a
+knot point and a vector of length `n+m`, and multiplication with a scalar, all of which will
+return a `StaticKnotPoint`.