feat(PPO_Bot): command abstraction layer — goto/aimTo controllers (#24)

- Add gotoTick/aimToTick controller functions (#25)
- Update network dims: actor 5→6, state 42→44 (#26)
- Rewrite mapActions for 6-dim command space (#27)
- Delete stale weight files (shape mismatch)
- Fix existing tests for new signatures

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

PPO_Bot/network.nim

@@ -3,6 +3,10 @@
 import arraymancer
 import std/[math, random]
 
+const
+  STATE_DIM*  = 44
+  ACTION_DIM* = 6
+
 type
   MLP* = object
     w1*, b1*: Tensor[float32]  # [hidden, input], [hidden]
@@ -12,7 +16,7 @@ type
   ActorCritic* = object
     actor*:  MLP
     critic*: MLP
-    logStd*: Tensor[float32]  # [5] — one per action dim
+    logStd*: Tensor[float32]  # [ACTION_DIM] — one per action dim
 
 proc initMLP*(inputDim, hiddenDim, outputDim: int): MLP =
   # Xavier/He-style init: scale weights by sqrt(2/fan_in)
@@ -24,9 +28,9 @@ proc initMLP*(inputDim, hiddenDim, outputDim: int): MLP =
   result.b3 = zeros[float32](outputDim)
 
 proc initActorCritic*(): ActorCritic =
-  result.actor  = initMLP(42, 64, 5)
-  result.critic = initMLP(42, 64, 1)
-  result.logStd = zeros[float32](5)  # init to 0 → std=1
+  result.actor  = initMLP(STATE_DIM, 64, ACTION_DIM)
+  result.critic = initMLP(STATE_DIM, 64, 1)
+  result.logStd = zeros[float32](ACTION_DIM)  # init to 0 → std=1
 
 proc forward*(mlp: MLP, x: Tensor[float32]): Tensor[float32] =
   ## x shape: [inputDim] (1D vector)
@@ -35,15 +39,15 @@ proc forward*(mlp: MLP, x: Tensor[float32]): Tensor[float32] =
   result  = mlp.w3 * h2 + mlp.b3
 
 proc actorForward*(ac: ActorCritic, state: Tensor[float32]): tuple[actions: Tensor[float32], logProb: float32] =
-  ## state: [42]. Returns sampled actions [5] and sum log-prob.
+  ## state: [STATE_DIM]. Returns sampled actions [ACTION_DIM] and sum log-prob.
   let mean = ac.actor.forward(state)
   # Floor logStd at -3 before exp → min std ≈ 0.05
   var clampedLogStd = ac.logStd.map(proc(v: float32): float32 = max(v, -3.0'f32))
   let std = clampedLogStd.map(proc(v: float32): float32 = exp(v))
 
-  var actions = newTensor[float32](5)
+  var actions = newTensor[float32](ACTION_DIM)
   var logP    = 0.0'f32
-  for i in 0..<5:
+  for i in 0..<ACTION_DIM:
     let mu = mean[i]
     let s  = std[i]
     let z  = gauss(0.0'f64, 1.0'f64).float32
@@ -55,7 +59,7 @@ proc actorForward*(ac: ActorCritic, state: Tensor[float32]): tuple[actions: Tens
   result = (actions: actions, logProb: logP)
 
 proc criticForward*(ac: ActorCritic, state: Tensor[float32]): float32 =
-  ## state: [42]. Returns scalar value estimate.
+  ## state: [STATE_DIM]. Returns scalar value estimate.
   let val = ac.critic.forward(state)
   result = val[0]
 
@@ -65,7 +69,7 @@ proc computeLogProb*(ac: ActorCritic, state, action: Tensor[float32]): float32 =
   let logStdClamped = ac.logStd.map(proc(v: float32): float32 = max(v, -3.0'f32))
   let std = logStdClamped.map(proc(v: float32): float32 = exp(v))
   var logP = 0.0'f32
-  for i in 0..<5:
+  for i in 0..<ACTION_DIM:
     let mu = mean[i]
     let s  = std[i]
     let diff = (action[i] - mu) / s