{"id":1976,"date":"2026-07-18T06:31:12","date_gmt":"2026-07-18T13:31:12","guid":{"rendered":"http:\/\/macdaddy4sure.ai\/?p=1976"},"modified":"2026-07-18T06:31:12","modified_gmt":"2026-07-18T13:31:12","slug":"driving-rl","status":"publish","type":"post","link":"http:\/\/macdaddy4sure.ai\/index.php\/2026\/07\/18\/driving-rl\/","title":{"rendered":"Driving RL"},"content":{"rendered":"\n

RL hub:<\/strong> ReinforcementLearning.md (ReinforcementLearning.md) \u00b7 neural policies:<\/p>\n\n\n\n

rl net enable driving \u2014 MathNN.md (MathNN.md).<\/p>\n\n\n\n

Driving RL is a simulation-first reinforcement-learning subsystem for lane<\/p>\n\n\n\n

keeping, lead-vehicle following, intersections, traffic controls, pedestrian<\/p>\n\n\n\n

yielding, bad-weather behavior, and emergency braking.<\/p>\n\n\n\n

Treat this as simulation\/review tooling unless a separate safety-certified<\/p>\n\n\n\n

control layer exists. Do not connect it directly to a real vehicle actuator.<\/p>\n\n\n\n

Quick Start<\/h2>\n\n\n\n

[text]
<\/em>driving rl status
driving rl features
driving rl actions
driving rl dry-run on
driving rl shield on
driving rl adapter set generic_driving
driving rl scenario lane_keep 120
driving rl scenario road_sign_stop 120
driving rl scenario speed_limit_sign 120
driving rl dashboard<\/p>\n\n\n\n

Training smoke run:<\/p>\n\n\n\n

[text]
<\/em>driving rl curriculum start sensor_smoke
driving rl scenario lead_follow 200
driving rl scenario stop_sign 200
driving rl scenario road_sign_stop 200
driving rl scenario speed_limit_sign 200
driving rl scenario traffic_light 200
driving rl scenario pedestrian_yield 200
driving rl checkpoint tag baseline<\/p>\n\n\n\n

Optional local DQN run:<\/p>\n\n\n\n

[text]
<\/em>driving rl dqn on
driving rl train
driving rl scenario road_sign_stop 300
driving rl scenario speed_limit_sign 300
driving rl dqn save checkpoints\/driving_road_sign_dqn.txt<\/p>\n\n\n\n

Modes<\/h2>\n\n\n\n

[text]
<\/em>off
assist
train
eval<\/p>\n\n\n\n

Commands:<\/p>\n\n\n\n

[text]
<\/em>driving rl on
driving rl assist
driving rl train
driving rl eval
driving rl off
driving rl status<\/p>\n\n\n\n

driving rl on enables driving and starts in assist mode with auto algorithm.<\/p>\n\n\n\n

Algorithms<\/h2>\n\n\n\n

[text]
<\/em>heuristic
qtable
dqn
auto<\/p>\n\n\n\n

Commands:<\/p>\n\n\n\n

[text]
<\/em>driving rl algorithm heuristic
driving rl algorithm qtable
driving rl algorithm dqn
driving rl algorithm auto
driving rl dqn on
driving rl dqn status
driving rl dqn save checkpoints\/driving_mathnn_dqn.txt
driving rl dqn load checkpoints\/driving_mathnn_dqn.txt
driving rl epsilon <value><\/p>\n\n\n\n

auto uses Q-table while training and heuristic behavior outside training. If<\/p>\n\n\n\n

the MathNN\/RLAgentNet driving policy is enabled, the neural policy can select<\/p>\n\n\n\n

actions through the existing driving policy slot. dqn enables the local<\/p>\n\n\n\n

MathNN-backed DQN learner for the driving agent and trains it from Driving RL<\/p>\n\n\n\n

transitions while mode is train.<\/p>\n\n\n\n

State Vector<\/h2>\n\n\n\n

DrivingState now has 32 normalized features:<\/p>\n\n\n\n

[text]
<\/em>s0   lane offset normalized around center
s1   time-to-collision risk
s2   collision probability
s3   vehicle velocity normalized
s4   lead vehicle distance normalized
s5   steering angle normalized
s6   left lane x normalized
s7   right lane x normalized
s8   lead relative velocity normalized
s9   track stable flag
s10  lead vehicle present flag
s11  lane width normalized
s12  lane confidence
s13  road curvature
s14  speed-limit ratio
s15  stop-sign flag
s16  traffic-light risk
s17  pedestrian risk
s18  intersection proximity
s19  route progress
s20  lane-departure\/offroad risk
s21  visibility risk
s22  jerk\/comfort cost
s23  safety context flag
s24  road-sign model confidence
s25  road-sign proximity
s26  road-sign stop\/yield requirement
s27  road-sign speed-limit present flag
s28  posted-speed overage
s29  road-sign warning\/no-entry\/crosswalk flag
s30  road-sign x position
s31  road-sign control signal<\/p>\n\n\n\n

The DrivingPerception input now supports lane confidence, road curvature,<\/p>\n\n\n\n

speed limits, route progress, intersection distance, traffic lights, stop signs,<\/p>\n\n\n\n

pedestrians, crosswalks, visibility, jerk, lane departure, offroad, and<\/p>\n\n\n\n

collision events in addition to the original lane\/lead vehicle fields. It also<\/p>\n\n\n\n

accepts trained road-sign detector output: label, confidence, bounding box,<\/p>\n\n\n\n

estimated distance, stop\/yield\/speed-limit\/warning\/no-entry flags, and posted<\/p>\n\n\n\n

speed limit.<\/p>\n\n\n\n

The driving camera loop passes the active object-detection model rows into<\/p>\n\n\n\n

Driving RL. Labels such as stop sign, stop_sign, yield_sign,<\/p>\n\n\n\n

speed_limit_35, traffic_light_red, traffic_light_yellow, wrong_way,<\/p>\n\n\n\n

school, crosswalk, and similar road-sign classes are normalized into the<\/p>\n\n\n\n

road-sign state features and safety shield.<\/p>\n\n\n\n

Actions<\/h2>\n\n\n\n

[text]
<\/em>0   maintain
1   accelerate
2   coast
3   brake_light
4   brake_moderate
5   brake_heavy
6   steer_left
7   steer_right
8   lane_center
9   yield
10  emergency_stop
11  crawl_forward<\/p>\n\n\n\n

Use:<\/p>\n\n\n\n

[text]
<\/em>driving rl actions<\/p>\n\n\n\n

Safety Shield<\/h2>\n\n\n\n

The safety shield wraps any selected policy action. It can override exploration,<\/p>\n\n\n\n

Q-table, heuristic, or neural choices when the perception state indicates:<\/p>\n\n\n\n

[text]
<\/em>collision event
critical lead vehicle risk
red light or stop sign near intersection
road-sign model stop\/yield\/no-entry signal
posted speed-limit overage
pedestrian near crosswalk
lane departure or offroad state<\/p>\n\n\n\n

Commands:<\/p>\n\n\n\n

[text]
<\/em>driving rl shield on
driving rl shield off
driving rl dry-run on
driving rl dry-run off
driving rl explain<\/p>\n\n\n\n

Keep dry-run on unless the recommendation is only being consumed by a<\/p>\n\n\n\n

simulation harness.<\/p>\n\n\n\n

Curriculum<\/h2>\n\n\n\n

Recommended phase order:<\/p>\n\n\n\n

[text]
<\/em>sensor_smoke
lane_keep
lead_follow
stop_sign
road_sign_stop
speed_limit_sign
traffic_light
pedestrian_yield
intersection
merge
bad_weather
emergency_brake
mixed_eval<\/p>\n\n\n\n

Commands:<\/p>\n\n\n\n

[text]
<\/em>driving rl curriculum status
driving rl curriculum start [phase]
driving rl curriculum next
driving rl curriculum phase <name><\/p>\n\n\n\n

Synthetic Scenarios<\/h2>\n\n\n\n

The scenario runner feeds synthetic DrivingPerception ticks through the normal<\/p>\n\n\n\n

DrivingRL pipeline. It is useful for fast smoke testing and checkpoint creation.<\/p>\n\n\n\n

[text]
<\/em>driving rl scenario list
driving rl scenario lane_keep 120
driving rl scenario lead_follow 200
driving rl scenario stop_sign 200
driving rl scenario road_sign_stop 200
driving rl scenario speed_limit_sign 200
driving rl scenario traffic_light 200
driving rl scenario pedestrian_yield 200
driving rl scenario bad_weather 200
driving rl scenario emergency_brake 200
driving rl scenario lane_departure 200
driving rl scenario merge 200<\/p>\n\n\n\n

These scenarios are not a substitute for a real simulator such as CARLA, Gazebo,<\/p>\n\n\n\n

or a custom physics sim, but they are good for verifying command paths, rewards,<\/p>\n\n\n\n

shield overrides, logging, and checkpoint behavior.<\/p>\n\n\n\n

Rewards And Feedback<\/h2>\n\n\n\n

The reward favors:<\/p>\n\n\n\n

    \n
  • lane centering,<\/li>\n\n\n\n
  • safe time-to-collision,<\/li>\n\n\n\n
  • reduced collision probability,<\/li>\n\n\n\n
  • obeying road-sign model stop\/yield signals,<\/li>\n\n\n\n
  • slowing when above a detected posted speed limit,<\/li>\n\n\n\n
  • route progress,<\/li>\n\n\n\n
  • recovery from lane-departure risk.<\/li>\n<\/ul>\n\n\n\n

    The reward penalizes:<\/p>\n\n\n\n

      \n
    • collision risk,<\/li>\n\n\n\n
    • pedestrian\/crosswalk risk,<\/li>\n\n\n\n
    • red-light\/stop-sign acceleration,<\/li>\n\n\n\n
    • accelerating through detected road-sign control signals,<\/li>\n\n\n\n
    • accelerating while above a detected posted speed limit,<\/li>\n\n\n\n
    • offroad\/lane departure,<\/li>\n\n\n\n
    • jerk\/comfort cost,<\/li>\n\n\n\n
    • unnecessary braking and throttle,<\/li>\n\n\n\n
    • safety shield overrides.<\/li>\n<\/ul>\n\n\n\n

      Manual feedback:<\/p>\n\n\n\n

      [text]
      <\/em>driving rl reward good [note]
      driving rl reward bad [note]
      driving rl reward <number> [note]<\/p>\n\n\n\n

      Replay, Dashboard, And Checkpoints<\/h2>\n\n\n\n

      Default directory:<\/p>\n\n\n\n

      [text]
      <\/em>D:\/driving_rl_logs<\/p>\n\n\n\n

      Files:<\/p>\n\n\n\n

      [text]
      <\/em>driving_rl_checkpoint.txt
      driving_rl_transitions.csv
      driving_rl_events.jsonl
      driving_rl_dashboard.json
      driving_rl_checkpoint_<tag>.txt<\/p>\n\n\n\n

      Commands:<\/p>\n\n\n\n

      [text]
      <\/em>driving rl dashboard
      driving rl replay summary [path]
      driving rl checkpoint tag <tag>
      driving rl checkpoint list
      driving rl checkpoint rollback <tag>
      driving rl save
      driving rl load
      driving rl save mysql
      driving rl load mysql<\/p>\n\n\n\n

      MySQL checkpoints use the configured driving_rl_checkpoint_profile.<\/p>\n\n\n\n

      Adapter Boundary<\/h2>\n\n\n\n

      Adapters should populate DrivingPerception from a simulator or perception<\/p>\n\n\n\n

      stack:<\/p>\n\n\n\n

      [text]
      <\/em>lane lines and confidence
      lead vehicle distance, velocity, TTC, collision probability
      ego speed, steering angle, jerk
      traffic light and stop sign state
      pedestrian\/crosswalk distance
      visibility\/weather\/night condition
      route progress and intersection distance
      lane departure\/offroad\/collision events<\/p>\n\n\n\n

      Commands:<\/p>\n\n\n\n

      [text]
      <\/em>driving rl adapter status
      driving rl adapter set <name>
      driving rl features<\/p>\n\n\n\n

      Lua Globals<\/h2>\n\n\n\n

      The subsystem exposes Driving RL helpers to Lua through<\/p>\n\n\n\n

      _DrivingRL::RegisterLuaGlobals:<\/p>\n\n\n\n

      [text]
      <\/em>driving_rl_status()
      driving_rl_set_mode(mode)
      driving_rl_scenario(name, ticks)
      driving_rl_shield(enabled)
      driving_rl_dashboard()<\/p>\n\n\n\n

      Safe Workflow<\/h2>\n\n\n\n
        \n
      1. Feed simulation perception only.<\/li>\n\n\n\n
      2. Start with driving rl dry-run on and driving rl shield on.<\/li>\n\n\n\n
      3. Run synthetic scenarios.<\/li>\n\n\n\n
      4. Inspect driving rl explain, driving rl dashboard, and replay summaries.<\/li>\n\n\n\n
      5. Train in a real simulator only after synthetic checks pass.<\/li>\n\n\n\n
      6. Save checkpoint tags before changing algorithms or epsilon.<\/li>\n\n\n\n
      7. Evaluate offline before using a learned policy as an assist recommendation.<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"

        RL hub: ReinforcementLearning.md (ReinforcementLearning.md) \u00b7 neural policies: rl net enable driving \u2014 MathNN.md (MathNN.md). Driving RL is a simulation-first reinforcement-learning subsystem for lane keeping, lead-vehicle following, intersections, traffic controls, pedestrian yielding, bad-weather behavior, and emergency braking. Treat this as simulation\/review tooling unless a separate safety-certified control layer exists. Do not connect it directly to a […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1976","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/posts\/1976","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/comments?post=1976"}],"version-history":[{"count":1,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/posts\/1976\/revisions"}],"predecessor-version":[{"id":1977,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/posts\/1976\/revisions\/1977"}],"wp:attachment":[{"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/media?parent=1976"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/categories?post=1976"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/macdaddy4sure.ai\/index.php\/wp-json\/wp\/v2\/tags?post=1976"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}