API tools faq
paste
Advertisement
alexhernandezroblox

Untitled

alexhernandezroblox
May 12th, 2025
30
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Lua 9.62 KB | None | 0 0
raw download clone embed print report
  1. if not game:GetService("RunService"):IsRunMode() and script.Name ~= "AIsDeepQClone" then
  2.     script.Enabled = false
  3. elseif game:GetService("RunService"):IsRunMode() then
  4.     script.AIShutdown.Value = true
  5. end
  6. if script.Disabled then return end
  7.  
  8. local RunService = game:GetService("RunService")
  9. local DataStoreService = game:GetService("DataStoreService")
  10. local NetworkStorage = DataStoreService:GetDataStore("NetworkStorage")
  11. local SS = game.ServerStorage
  12. local OpenML = require(SS.OpenML)
  13.  
  14. local function sigmoid(x)
  15.     return 1 / (1 + math.exp(-x))
  16. end
  17.  
  18. local ITERATION_GUI = workspace.Iteration.SurfaceGui.TextLabel
  19. local FINISHES_GUI = workspace.Finishes.SurfaceGui.TextLabel
  20. local GOAL = workspace.goal
  21. local SPAWN = workspace.spawn
  22. local PASSED_COORDS = {
  23.     workspace.Passed1.Position.X, workspace.Passed2.Position.X,
  24.     workspace.Passed3.Position.X, workspace.Passed4.Position.X,
  25.     workspace.Passed5.Position.X, workspace.Passed6.Position.X,
  26.     workspace.Passed7.Position.X, workspace.Passed8.Position.X,
  27.     workspace.Passed9.Position.X, workspace.Passed10.Position.X,
  28.     workspace.Passed11.Position.X, workspace.Passed12.Position.X,
  29.     workspace.Passed13.Position.X, workspace.Passed14.Position.X,
  30.     workspace.Passed15.Position.X, workspace.Passed16.Position.X,
  31. }
  32.  
  33. FINISHES_GUI.FinishCount.Value = 0
  34. FINISHES_GUI.Text = "Finishes: 0"
  35. ITERATION_GUI.IterationCount.Value = 0
  36. ITERATION_GUI.Text = "Iteration: 0"
  37. workspace.Finishes.CFrame = CFrame.new(-128.949203, 90.3132095, 84.4396667, 0.999922872, 0.00161167711, -0.0123128593, 0, 0.991541922, 0.129786715, 0.0124178901, -0.129776701, 0.991465509)
  38. workspace.Iteration.CFrame = CFrame.new(-128.988525, 86.216301, 87.6062622, 0.999922872, 0.00161167711, -0.0123128593, 0, 0.991541922, 0.129786715, 0.0124178901, -0.129776701, 0.991465509)
  39.  
  40. local AGENT_BODY = SS.AgentRig:Clone()
  41. script.Agent.Value = AGENT_BODY
  42. local HUMANOID_ROOT = AGENT_BODY.PrimaryPart
  43. local HUMANOID = AGENT_BODY.Humanoid
  44.  
  45. local ADAM_OPTIMIZER = OpenML.Optimizers.Adam.new()
  46. local PROPAGATOR = OpenML.Algorithms.Propagator
  47. local FUNCTION_ACTIVATION = OpenML.ActivationFunctions.ReLU
  48. local FUNCTION_LOSS = OpenML.Calculate.MSE
  49. local DQL = OpenML.Algorithms.DQL.new()
  50. local NETWORK
  51. local NetworkIterationsSave, NetworkFinishesSave, steps = nil, nil, 0
  52. local consecutiveFalls = 0
  53. local livingPenalty = 0.002
  54.  
  55. if script.AIUpdated.Value then
  56.     local data1 = NetworkStorage:GetAsync("EfficientNetwork1")
  57.     NetworkIterationsSave = NetworkStorage:GetAsync("EfficientNetwork2")
  58.     NetworkFinishesSave = NetworkStorage:GetAsync("EfficientNetwork3")
  59.     steps = NetworkStorage:GetAsync("EfficientNetworkSteps")
  60.     NETWORK = OpenML.Resources.MLP.Decompress(data1, 'IEEE754')
  61.     local savedCFrame = NetworkStorage:GetAsync("EfficientNetworkCFrame")
  62.     if savedCFrame then
  63.         HUMANOID_ROOT.CFrame = CFrame.new(table.unpack(savedCFrame))
  64.     end
  65.  
  66. elseif script.AIShutdown.Value then
  67.     local data2 = NetworkStorage:GetAsync("ShutdownNetwork1")
  68.     NetworkIterationsSave = NetworkStorage:GetAsync("ShutdownNetwork2")
  69.     NetworkFinishesSave = NetworkStorage:GetAsync("ShutdownNetwork3")
  70.     steps = NetworkStorage:GetAsync("ShutdownNetworkSteps")
  71.     NETWORK = OpenML.Resources.MLP.Decompress(data2, 'IEEE754')
  72.     local savedCFrame = NetworkStorage:GetAsync("ShutdownNetworkCFrame")
  73.     if savedCFrame then
  74.         HUMANOID_ROOT.CFrame = CFrame.new(table.unpack(savedCFrame))
  75.     end
  76.  
  77. else
  78.     NETWORK = OpenML.Resources.MLP.new({7, 32, 32, 16, 2}, function()
  79.         return Random.new():NextNumber() * 2 - 1
  80.     end)
  81.     HUMANOID_ROOT.CFrame = SPAWN.CFrame
  82. end
  83.  
  84. setmetatable(NETWORK, { __index = PROPAGATOR })
  85.  
  86. if NetworkIterationsSave and NetworkFinishesSave then
  87.     ITERATION_GUI.IterationCount.Value = NetworkIterationsSave
  88.     ITERATION_GUI.Text = "Iteration: " .. tostring(NetworkIterationsSave)
  89.     FINISHES_GUI.FinishCount.Value = NetworkFinishesSave
  90.     FINISHES_GUI.Text = "Finishes: " .. tostring(NetworkFinishesSave)
  91.     PREVIOUS_DIST = (HUMANOID_ROOT.Position - GOAL.Position).Magnitude
  92. end
  93.  
  94. AGENT_BODY.Parent = workspace
  95. HUMANOID.AutoRotate = false
  96.  
  97. DQL.OnForwardPropagation = function(states)
  98.     return NETWORK:ForwardPropagation(states, FUNCTION_ACTIVATION)
  99. end
  100.  
  101. local LEARNING_RATE = 0.001
  102. DQL.OnBackPropagation = function(activations, target)
  103.     return NETWORK:BackPropagation(activations, target, {
  104.         ActivationFunction = FUNCTION_ACTIVATION,
  105.         Optimizer = ADAM_OPTIMIZER,
  106.         LearningRate = LEARNING_RATE
  107.     })
  108. end
  109.  
  110. local Params = RaycastParams.new()
  111. Params.FilterDescendantsInstances = {AGENT_BODY}
  112. Params.FilterType = Enum.RaycastFilterType.Exclude
  113.  
  114. local PREVIOUS_DIST = (SPAWN.Position - GOAL.Position).Magnitude
  115. local LAST_PASSED = 0
  116. local SECOND_LAST_PASSED = 0 -- before LAST_PASSED
  117. local SAME_PASSED = 0 -- has the agent passed the same platform for 2 episodes/iterations? e.g. if he passed platform 5 ONLY and EXACTLY twice in a row then this val would be 2
  118. local TOTAL_REWARD = 0
  119. local TIME_STILL = 0
  120.  
  121. game:BindToClose(function()
  122.     warn("shutting down")
  123.     local compressed = OpenML.Resources.MLP.Compress(NETWORK, "IEEE754")
  124.     NetworkStorage:SetAsync("ShutdownNetwork1", compressed)
  125.     NetworkStorage:SetAsync("ShutdownNetwork2", ITERATION_GUI.IterationCount.Value)
  126.     NetworkStorage:SetAsync("ShutdownNetwork3", FINISHES_GUI.FinishCount.Value)
  127.     NetworkStorage:SetAsync("ShutdownNetworkCFrame", {HUMANOID_ROOT.CFrame:GetComponents()})
  128.     NetworkStorage:SetAsync("ShutdownNetworkSteps", steps)
  129.     warn("saved")
  130. end)
  131.  
  132. RunService.Stepped:Connect(function(_, deltaTime)
  133.     steps += 1
  134.  
  135.     local distToGoal = (HUMANOID_ROOT.Position - GOAL.Position).Magnitude
  136.     local downRay = workspace:Raycast(HUMANOID_ROOT.Position, Vector3.new(0, -1, 0) * 4, Params)
  137.     local furthestRay = workspace:Raycast(HUMANOID_ROOT.Position, Vector3.new(-1, -1, 0) * 4, Params)
  138.     local intermediateRay = workspace:Raycast(HUMANOID_ROOT.Position, Vector3.new(-0.5, -1, 0) * 4, Params)
  139.  
  140.     local function getDepth(ray)
  141.         return ray and (ray.Position - HUMANOID_ROOT.Position).Magnitude or 4
  142.     end
  143.  
  144.     local inputs = {
  145.         distToGoal,
  146.         (downRay and downRay.Instance) and 1 or 0,
  147.         HUMANOID_ROOT.Velocity.Magnitude,
  148.         HUMANOID_ROOT.Velocity.Y,
  149.         getDepth(downRay),
  150.         getDepth(furthestRay),
  151.         getDepth(intermediateRay),
  152.     }
  153.  
  154.     local activations = NETWORK:ForwardPropagation(inputs, FUNCTION_ACTIVATION)
  155.     local actions = activations[#activations]
  156.     local epsilon_start = 1
  157.     local epsilon_end = 0.05
  158.     local decay_steps = 2e6
  159.     local epsilon = math.max(epsilon_end, epsilon_start - (steps / decay_steps) * (epsilon_start - epsilon_end))
  160.     local isEpsilon = math.random() <= epsilon
  161.     local chosenAction = isEpsilon and math.random(1, 2) or (actions[1] > actions[2] and 1 or 2)
  162.  
  163.     if chosenAction == 1 then
  164.         HUMANOID:MoveTo(GOAL.Position)
  165.     else
  166.         HUMANOID:MoveTo(HUMANOID_ROOT.Position)
  167.         HUMANOID.Jump = true
  168.     end
  169.  
  170.     local reward
  171.     local distDiff = PREVIOUS_DIST - distToGoal
  172.     if distToGoal < PREVIOUS_DIST and math.abs(distDiff) > 0.26 then
  173.         reward = distDiff * 0.3
  174.         TIME_STILL = 0
  175.     else
  176.         reward = 0
  177.         TIME_STILL += deltaTime
  178.         warn("TIME_STILL:", TIME_STILL)
  179.         if TIME_STILL > 3 then
  180.             warn("Punishing")
  181.             reward = -0.1 + ((TIME_STILL - 3) * -0.01) -- reward punishment calculation in milliseconds
  182.         end
  183.     end
  184.  
  185.     local xPos = HUMANOID_ROOT.Position.X
  186.     for i, threshold in ipairs(PASSED_COORDS) do
  187.         if xPos < threshold and LAST_PASSED < i then
  188.             LAST_PASSED = i
  189.             reward = i * 20
  190.             break
  191.         end
  192.     end
  193.     TOTAL_REWARD += reward
  194.     TOTAL_REWARD -= livingPenalty
  195.  
  196.     local touchingBottom = downRay and downRay.Instance == workspace.bottom
  197.     if touchingBottom then
  198.         local fallPenalty = -10
  199.         if (TIME_STILL - 3) * -0.02 >= fallPenalty then
  200.             consecutiveFalls += 1
  201.         else
  202.             consecutiveFalls = 0
  203.         end
  204.         if SECOND_LAST_PASSED == LAST_PASSED and SECOND_LAST_PASSED > 0 then
  205.             SAME_PASSED += 1
  206.         else
  207.             SAME_PASSED = 0
  208.         end
  209.         SECOND_LAST_PASSED = LAST_PASSED
  210.         if SAME_PASSED >= 5 then
  211.             fallPenalty = reward + 5
  212.         elseif consecutiveFalls >= 10 then
  213.             fallPenalty = ((TIME_STILL - 3) * -0.02) + 1
  214.         end
  215.         reward = fallPenalty
  216.         TOTAL_REWARD = 0
  217.         TIME_STILL = 0
  218.         LAST_PASSED = 0
  219.         HUMANOID_ROOT.CFrame = SPAWN.CFrame
  220.         ITERATION_GUI.IterationCount.Value += 1
  221.         warn("line 227 + 1")
  222.         ITERATION_GUI.Text = "Iteration: " .. tostring(ITERATION_GUI.IterationCount.Value)
  223.         PREVIOUS_DIST = (HUMANOID_ROOT.Position - GOAL.Position).Magnitude
  224.     end
  225.  
  226.     PREVIOUS_DIST = distToGoal
  227.     print("Reward:",reward - livingPenalty)
  228.     print("Total reward:",TOTAL_REWARD)
  229.     print("State:", unpack(actions))
  230.     print("Current Epsilon:",epsilon)
  231.     print("Selected Action:", if chosenAction == 1 then "Move" else "Jump", "Epsilon:", isEpsilon)
  232.  
  233.     if distToGoal < 2 then
  234.         if HUMANOID_ROOT.CFrame ~= SPAWN.CFrame then
  235.             reward = 480 -- finishing obby reward is equal to passed16 reward * 1.5
  236.         end
  237.  
  238.         warn("Finished Simulation!")
  239.         HUMANOID_ROOT.CFrame = SPAWN.CFrame
  240.         ITERATION_GUI.IterationCount.Value += 1
  241.         warn('line 247 + 1')
  242.         ITERATION_GUI.Text = "Iteration: " .. tostring(ITERATION_GUI.IterationCount.Value)
  243.         FINISHES_GUI.FinishCount.Value += 1
  244.         FINISHES_GUI.Text = "Finishes: " .. tostring(FINISHES_GUI.FinishCount.Value)
  245.         PREVIOUS_DIST = (HUMANOID_ROOT.Position - GOAL.Position).Magnitude
  246.  
  247.         local compressed = OpenML.Resources.MLP.Compress(NETWORK, 'IEEE754')
  248.         NetworkStorage:SetAsync("EfficientNetwork1", compressed)
  249.         NetworkStorage:SetAsync("EfficientNetwork2", ITERATION_GUI.IterationCount.Value)
  250.         NetworkStorage:SetAsync("EfficientNetwork3", FINISHES_GUI.FinishCount.Value)
  251.         NetworkStorage:SetAsync("EfficientNetworkCFrame", {HUMANOID_ROOT.CFrame:GetComponents()})
  252.         NetworkStorage:SetAsync("EfficientNetworkSteps", steps)
  253.     end
  254.  
  255.     DQL:Learn({
  256.         State  = inputs,
  257.         Action = chosenAction,
  258.         Reward = reward - livingPenalty,
  259.     })
  260. end)
  261.  
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!