Advanced CARLA Driving Simulator
Budget: $750 – $1,500 USD
I want to build a full–featured simulator on top of the CARLA open-source platform that serves two intertwined goals: practical driving training and rigorous human-factors research. The same codebase should let an instructor walk a learner through realistic city streets, high-speed expressways, or rugged off-road paths, and then switch seamlessly into experimental modes where we can study how stress, drowsiness, or other cognitive loads influence driving behaviour. Need for Left hand as well as right hand driving
Core requirements
• Scenarios: Ready-made environments for urban traffic, multilane highways, and unpaved terrain, each populated with dynamic actors (vehicles, pedestrians, cyclists) and weather/time-of-day variations that I can tweak from a config file or Python script.
• Vehicle dynamics: Acceleration & braking and steering & cornering must feel authentic and respect real-world limits; if you have a physics plug-in you prefer, just make sure it slots cleanly into CARLA’s API.
• Research layer: Hooks that let me trigger events (e.g., sudden obstacle, reduced visibility, sleepiness, distractions, alcohol etc) and log driver inputs, biometrics, and vehicle telemetry at millisecond precision. I’ll integrate my own heart-rate and eye-tracking devices, so expose a clean TCP/UDP or REST interface for data sync.
• Training layer: A user-friendly dashboard with admin panel, where an instructor or student selects scenario, sets objectives, and reviews trainee performance metrics (speed profile, lane keeping, reaction time).
Deliverables
1. Source code in a private Git repository with clear commit history.
2. A one-click installer or Docker image for Windows and Linux.
3. At least seven demo scenarios per driving context (urban, highway, off-road, hills, night, rain, peak traffic hours).
4. Documentation: setup guide, API reference, and sample research scripts.
5. Short video walkthrough proving dynamics fidelity and scenario switching.
Acceptance criteria
• Running the installer on a clean machine launches CARLA with all assets
• Speeds, accelerations, and steering angles match real vehicle data within ±5 %.
• Scenario reload takes ≤10 s and never crashes the engine.
• Logged data streams have timestamp drift <2 ms over a 30-minute session.
If you have prior CARLA mods, reinforcement-learning projects, or published human-factors work, please share a link so I can evaluate fit quickly. I’m ready to start as soon as we agree on a milestone plan and timeline.
Core requirements
• Scenarios: Ready-made environments for urban traffic, multilane highways, and unpaved terrain, each populated with dynamic actors (vehicles, pedestrians, cyclists) and weather/time-of-day variations that I can tweak from a config file or Python script.
• Vehicle dynamics: Acceleration & braking and steering & cornering must feel authentic and respect real-world limits; if you have a physics plug-in you prefer, just make sure it slots cleanly into CARLA’s API.
• Research layer: Hooks that let me trigger events (e.g., sudden obstacle, reduced visibility, sleepiness, distractions, alcohol etc) and log driver inputs, biometrics, and vehicle telemetry at millisecond precision. I’ll integrate my own heart-rate and eye-tracking devices, so expose a clean TCP/UDP or REST interface for data sync.
• Training layer: A user-friendly dashboard with admin panel, where an instructor or student selects scenario, sets objectives, and reviews trainee performance metrics (speed profile, lane keeping, reaction time).
Deliverables
1. Source code in a private Git repository with clear commit history.
2. A one-click installer or Docker image for Windows and Linux.
3. At least seven demo scenarios per driving context (urban, highway, off-road, hills, night, rain, peak traffic hours).
4. Documentation: setup guide, API reference, and sample research scripts.
5. Short video walkthrough proving dynamics fidelity and scenario switching.
Acceptance criteria
• Running the installer on a clean machine launches CARLA with all assets
• Speeds, accelerations, and steering angles match real vehicle data within ±5 %.
• Scenario reload takes ≤10 s and never crashes the engine.
• Logged data streams have timestamp drift <2 ms over a 30-minute session.
If you have prior CARLA mods, reinforcement-learning projects, or published human-factors work, please share a link so I can evaluate fit quickly. I’m ready to start as soon as we agree on a milestone plan and timeline.