New Systems

What We Did

· Lane Commitment Logic:

· We introduced a system where, once an AI car selects an open lane, it “commits” to that lane for a minimum amount of time (or until it becomes blocked), instead of constantly re-evaluating and switching lanes every frame.

· Safe Lane Selection:

· We added logic to ensure the AI only commits to a lane if it is both open and safe (not too close to the road edge or hazards like telephone poles, hydrants, or parked cars).

· This uses a buffer check and a dedicated hazard layer mask.

· Bounds Checking:

· We added checks to prevent the AI from trying to steer to an invalid lane index, which prevents errors and ensures the car always has a valid target.

---

Why We Did It

· Reduce Erratic Steering:

Previously, the AI would constantly try to return to the center or switch lanes too frequently, causing unnatural “back and forth” steering, especially when open lanes were available.

· Prevent Collisions with Roadside Hazards:

Without awareness of the road edge, AI cars could select far left or right lanes and hit objects on the sidewalk or curb.

· Increase Realism and Smoothness:

Real drivers (and race cars) commit to a lane and only change when necessary, especially when a lane is clear and safe.

___________________________________________________________________________________________________________________________________________________________________________________________________________

How It Works

· Lane Commitment:

· When the AI finds an open and safe lane, it sets a timer and “commits” to that lane.

· It will stay in that lane for a set duration, unless the lane becomes blocked.

· Only after the timer expires or the lane is blocked will the AI reconsider changing lanes.

· Safe Lane Check:

· Before committing to a lane (especially the outer lanes), the AI checks for hazards using raycasts or spherecasts to the side and ahead.

· Only lanes that are both open (no traffic) and safe (no hazards) are considered.

· Bounds Checking:

· The AI always checks that the lane index is valid before steering, preventing crashes and fallback to a safe lane if needed.

---

Benefits

· Smoother, More Realistic Driving:

AI cars now drive in a chosen lane for a realistic amount of time, just like human drivers, instead of jittering between lanes.

· Fewer Collisions:

By avoiding unsafe lanes near the road edge, AI cars are much less likely to hit telephone poles, hydrants, or parked cars.

· Better Traffic Flow:

AI cars make smarter, less frequent lane changes, leading to more believable and efficient traffic movement.

· Easier Tuning and Maintenance:

The code is modular, with clear helper methods and safety checks, making it easier to adjust and extend in the future.

---

In summary:

These changes make your AI cars drive more like skilled, cautious human drivers—committing to open lanes, avoiding hazards, and steering smoothly—resulting in a more fun, realistic, and robust city racing experience.

What We Changed

· Persistent Learning Data Structure:

· Added serializable classes (LearningData, VehicleLearningData, PassData) to store AI learning data (track, vehicle, passes, etc.).

· Save/Load Logic:

· Implemented methods to save learning data to disk after a race and load it at startup if enabled.

· Inspector Checkbox:

· Added a checkbox in the Unity Inspector to let users enable or disable saving learning data.

· Naming Conflict Fix:

· Renamed the serializable learning data field to avoid a naming conflict with the in-memory learning dictionary.

---

Why This Is Good

· AI Progression:

The AI now remembers what it learned from previous races, so it gets better over time instead of starting from scratch each session.

· User Control:

The checkbox gives users the choice to save or not, respecting privacy and disk usage.

· Performance:

Data is only saved after the race, not during, so there’s no performance hit while racing.

· Maintainability:

Clear separation between runtime learning (in-memory) and persistent storage (on disk) makes the code easier to understand and extend.

---

What’s Different Now?

· Before:

The AI would forget all learning after each session—no persistent improvement.

· Now:

The AI can load previous learning data and continue improving, making it more realistic and competitive over time. Users can opt in or out of this feature.