How ezharjan/autoscenegen works
AutoSceneGen is a research toolkit for turning traffic descriptions (and, in a separate image workflow, images) into simulator scenarios, collecting synthetic trajectories, and preparing datasets for motion-planning or prediction experiments. The main documented path is description → LLM-generated scenario script → CARLA execution → trajectory collection → dataset use. The excerpts show the prompt, generation, validation, and saving connections; the README documents broader simulator and dataset capabilities. Collector internals and several conversion scripts were not sampled, so their links are limited to documented relationships or omitted.
Main entry points: gen_scenario_by_llm.py Single generation · batch_gen_scenario_by_llm.py Batch generation · save_content.py Validate and save · img2txt.py Image description
Components 25
Scenario Authoring
- Prompt examples
Codes/ICL/questions.py - Single generation
Codes/ICL/gen_scenario_by_llm.py - Batch generation
Codes/ICL/batch_gen_scenario_by_llm.py - Validate and save
Codes/ICL/save_content.py - Script validation
Codes/ICL/validate_content.py
CARLA Simulation
- Scenario scripts
Codes/CarlaPy/scenarios/0315203222.py - Manual control
Codes/CarlaPy/manual_control.py
Data Collection
- Apollo collection
CarlaDatasetCollector/collect_apollo_data.py - Pishgu collection
CarlaDatasetCollector/collect_pishgu_data.py - Automated collection
CarlaDatasetCollector/auto_collect_data.py - Trajectory data
Dataset Preparation
- Log conversion
CarlaDatasetCollector/log2csv.py - CSV data conversion
CarlaDatasetCollector/csv2data.py - Data to CSV
CarlaDatasetCollector/data2csv.py - MSRL conversion
Dataset/MSRL-Data&Toolkits/pishgu2msrl.py - Pedestrian conversion
Dataset/Pishgu-All-Pedestrian(NGSIM-style)/csvs2dataset.py
Image Inputs
- Image prompts
Codes/ICL/questions.py - Image description
Codes/ICL/img2txt.py - Batch image extraction
Codes/ICL/extract_info_from_imgs.py
Other components
- Scenario author
- OpenAI GPT-4
- CARLA simulator
- Planner training
- Image analyst
- Gemini Vision
Connections 30
- Scenario author → Single generation: describes scenario
README.md - Single generation → Prompt examples: uses prompts
Codes/ICL/gen_scenario_by_llm.py - Single generation → OpenAI GPT-4: requests generation
Codes/ICL/gen_scenario_by_llm.py - Single generation → Validate and save: saves response
Codes/ICL/gen_scenario_by_llm.py - Scenario author → Batch generation: supplies scenarios
README.md - Batch generation → Prompt examples: uses prompts
Codes/ICL/batch_gen_scenario_by_llm.py - Batch generation → OpenAI GPT-4: requests generation
Codes/ICL/batch_gen_scenario_by_llm.py - Batch generation → Validate and save: saves responses
Codes/ICL/batch_gen_scenario_by_llm.py - Validate and save → Script validation: validates code
Codes/ICL/save_content.py - Validate and save → Scenario scripts: writes scripts
Codes/ICL/save_content.py - Scenario scripts → CARLA simulator: executes scenarios
README.md - Manual control → CARLA simulator: controls simulation
Codes/CarlaPy/manual_control.py - CARLA simulator → Apollo collection: provides simulated data
- CARLA simulator → Pishgu collection: provides simulated data
- CARLA simulator → Automated collection: provides simulated data
- Apollo collection → Trajectory data: collects trajectories
README.md - Pishgu collection → Trajectory data: collects trajectories
README.md - Automated collection → Trajectory data: collects data
- Trajectory data → Planner training: trains or evaluates
README.md - Trajectory data → Log conversion: converts logs
- Trajectory data → CSV data conversion: converts data
- Trajectory data → Data to CSV: exports CSV
- Trajectory data → MSRL conversion: prepares MSRL data
- Trajectory data → Pedestrian conversion: prepares pedestrian data
- Image analyst → Image description: provides image
Codes/ICL/img2txt.py - Image description → Image prompts: selects prompt
Codes/ICL/img2txt.py - Image description → Gemini Vision: requests description
Codes/ICL/img2txt.py - Image analyst → Batch image extraction: provides images
Codes/ICL/extract_info_from_imgs.py - Batch image extraction → Image prompts: selects prompt
Codes/ICL/extract_info_from_imgs.py - Batch image extraction → Gemini Vision: requests extraction
Codes/ICL/extract_info_from_imgs.py