born5149/autonomous-vehicle-sensor-fusion

Autonomous Vehicle Raw Sensor Telemetry & Simulation Dataset

This repository contains raw, uncompressed sensor buffers, massive neural network weights, federated learning node dumps, and VRAM memory snapshots collected from autonomous vehicle test fleets. Data is provided "as is" for offline perception model training, system debugging, and Hardware-in-the-Loop (HIL) simulations.

Dataset Structure (Full Schema)

Due to horizontal scaling and massive daily ingestions, files are strictly indexed by sensor hardware or software channel (_chX_). Cumulative data batches or continuous streams exceeding 20-50GB are sharded into sequential volumes (_part1, _part2, etc.) to ensure data integrity during upload.

1. Core Memory, Compute Nodes & System Logic

• autonomous_core_memory_* - Core autonomous driving kernel crash dumps and critical state memory snapshots.
• sensor_fusion_init_dump_* - Initial state vectors and calibration matrices loaded during cold boot.
• kernel_panic_crashdump_full_* - Complete system state dumps triggered during catastrophic OS kernel failures.
• nvme_swap_cache_partition_* - High-speed NVMe swap partition dumps for memory leak analysis.
• deep_learning_accelerator_vram_dump_* - Raw VRAM dumps from onboard deep learning accelerators.
• autonomous_compute_node_nvme_swap_* - Swap memory snapshots from secondary compute nodes.
• edge_compute_node_memory_snapshot_* - Memory states from low-latency edge computing modules.
• fail_operational_state_memory_dump_* - Memory dumps captured during fail-safe operational handovers.
• nvidia_orin_gpu_memory_dump_* - Complete VRAM snapshots from the primary NVIDIA Orin compute nodes.

2. Spatial, Lidar & Radar Systems

• navigation_hd_map_cache_* - Pre-compiled High-Definition (HD) route map caches and localization tiles.
• hd_map_tile_index_* - Indexed cache volumes for high-density spatial routing.
• lidar_128ch_raw_spin_* - Raw 128-channel Lidar spin returns and intensity values.
• pointcloud_accumulation_buffer_* - Aggregated 3D spatial returns for dense pointcloud generation.
• occupancy_grid_3d_volume_* - 3D voxel occupancy grids for volumetric spatial mapping.
• solid_state_lidar_matrix_* - Non-mechanical solid-state Lidar returns for short-range obstacle mapping.
• fmcw_radar_doppler_datacube_* - Frequency Modulated Continuous Wave radar doppler data cubes.
• lidar_time_of_flight_raw_returns_* - Unprocessed time-of-flight measurements for spatial correction.
• millimeter_wave_radar_pointcloud_* - Raw mmWave radar returns for adverse weather penetration testing.

3. Optical, Vision & Infrared Sensors

• surround_cam_uncompressed_* - Raw, uncompressed optical feeds (.raw) from the 360-degree camera matrix.
• fisheye_surround_raw_* - Distorted, raw frame sequences from the parking camera matrix (.raw).
• surround_view_fisheye_tensor_cache_* - Processed tensor caches generated from the fisheye optical arrays.
• stereo_depth_map_* - High-resolution, unprocessed disparity maps from the forward stereo vision system.
• night_vision_raw_feed_* - Uncompressed infrared optical streams for low-light environment perception.
• thermal_ir_array_* - Uncalibrated 16-bit thermal sensor matrices for pedestrian and animal detection.
• front_telephoto_raw_stream_* - Long-range optical feeds (.raw) for high-speed object detection.
• hdr_bracket_sequence_* - High Dynamic Range exposure sequences for low-light sensor evaluation.
• stereo_disparity_volume_* - 3D disparity volumes calculated from stereo baseline calibration.
• uncompressed_yuv422_camera_streams_* - Heavy YUV422 video streams for ISP (Image Signal Processor) tuning.
• dense_optical_flow_video_tensors_* - Optical flow calculations extracted from continuous motion video.
• stereo_vision_disparity_cost_volumes_* - Disparity cost volumes for offline depth estimation optimization.
• event_camera_neuromorphic_stream_* - High-speed, microsecond-resolution event-based camera data streams.
• hyperspectral_scan_* - Multi-band hyperspectral imaging returns for advanced material classification.

4. Simulation, Render Caches & Digital Twins

• carla_unreal_engine_render_cache_* - Heavy, uncompressed rendering caches from Carla (Unreal Engine 5) simulations.
• synthetic_sim_environment_* - Virtual environment state captures and simulated object behaviors.
• synthetic_dataset_generation_batch_* - Massively generated synthetic edge-case batches for offline training.
• endurance_test_raw_accumulation_* - Continuous logging buffers generated during multi-hour endurance simulations.
• carla_sim_weather_rendering_cache_* - Physics-based weather simulation (rain/fog) rendering caches.
• unreal_engine_synthetic_rgbd_frames_* - Combined RGB and Depth synthetic frame dumps.
• scenario_runner_monte_carlo_rollouts_* - Stochastic scenario evaluations and crash-test boundary rollouts.
• hardware_in_the_loop_vram_dump_* - Memory captures during hardware-in-the-loop stress testing.
• digital_twin_physics_engine_state_* - Complete state arrays from the digital twin physics calculation engine.
• sensor_degradation_simulation_cache_* - Synthetic injection of sensor failures (lens flare, mud, packet loss).
• multi_agent_interaction_rollouts_* - Behavioral simulation caches involving multiple autonomous agents.
• neural_planner_replay_buffer_* - Action-state pairs for reinforcement learning and trajectory planning.

5. AI Weights, Tensors & Federated Learning

• transformer_attention_weights_* - Massive matrix weights and intermediate states from perception Transformer models.
• perception_tensor_activations_* - Intermediate tensor activation states for deep learning model debugging.
• model_training_epoch_checkpoint_* - Massive model weights preserved between continuous training epochs.
• federated_learning_node_dump_* - Weight updates and gradient dumps transferred from remote fleet nodes.
• sensor_fusion_matrix_* - Multidimensional state correlation matrices from the centralized perception engine.
• bev_features_tensor_cache_* - Bird's Eye View (BEV) feature maps extracted from multi-camera rigs.
• model_checkpoint_epoch_raw_* - Uncompressed epoch checkpoints captured during continuous cloud training.
• bev_transformer_attention_maps_* - Attention map tensors from Bird's Eye View projection networks.
• multimodal_fusion_network_activations_* - Activation tensors from LiDAR-Camera fusion bottlenecks.
• occupancy_network_voxel_grids_* - Uncompressed spatial volume tensors for object occupancy networks.
• weather_noise_filtering_activations_* - Network states from the deep weather de-noising architecture.
• point_pillar_feature_encoding_cache_* - Cached PointPillar features to accelerate LiDAR network retraining.

6. Network Telemetry, V2X & Continuous Integration

• rosbag2_raw_system_telemetry_* - High-frequency ROS2 system telemetry streams and asynchronous event logs.
• ros2_middleware_trace_* - Latency and execution trace logs from the ROS2 middleware communication layer.
• telemetry_baseband_dump_* - Raw 5G/V2X baseband radio frequency (RF) dumps for network latency analysis.
• fleet_unfiltered_sensor_log_* - Continuous, unfiltered multiplexed logs dumped directly from production fleets.
• continuous_integration_build_cache_* - Heavy binary caches used by CI/CD pipelines for onboard software deployment.
• pcap_ethernet_backbone_raw_* - Raw PCAP packet captures from the vehicle's internal gigabit ethernet backbone.
• v2x_5g_baseband_iq_samples_* - Baseband I/Q sample streams for V2X (Vehicle-to-Everything) connectivity testing.
• gigabit_ethernet_backbone_pcap_* - High-throughput intra-vehicle network traffic captures.
• fleet_telemetry_aggregation_buffer_* - Buffered synchronization chunks before cloud offloading.
• autonomous_fleet_shadow_mode_logs_* - Telemetry captures from fleet vehicles running perception models in shadow mode.
• time_synchronized_sensor_payload_* - PTP-synchronized sensor datagrams for timing jitter analysis.

---

Data Collection Status & Changelog

Current Phase: Phase 3 (Distributed Simulation & High-Density Traffic Logging) System Status: [ACTIVE] Continuous data ingestion in progress.

• [Recent Updates]: Phased out legacy unstructured ECU dumps. Transitioned to high-fidelity continuous Lidar (128-channel) and heavy Unreal Engine 5 simulation caches to improve offline training accuracy.
• [Architecture Note]: To accommodate horizontal scaling, files are strictly indexed by sensor hardware or software channel (_chX_). Extremely large dumps or cumulative data batches (exceeding 10-20GB) generated during high-density traffic events are sharded into sequential volumes (_part1, _part2) to prevent network timeout during uploads and ensure data integrity.
• [Upload Pipeline Queue]: Advanced sensor categories and massive simulation batches will be synced to this repository over the next several weeks to comply with bandwidth allocation limits and cloud ingest quotas.

Note to engineering team: File sizes vary significantly — ranging from small initialization matrices (~10MB) to massive raw simulation chunks, audio accumulation batches, and transformer weights (up to 50GB+). Do not truncate or compress these files during local parsing to preserve cryptographically verified entropy. Treat all .dat and .raw formats as cryptographically rigid bitstreams. The parsing layer is intentionally disabled (viewer: false) to preserve binary integrity.