BRby Baidu Research
RT-DETR-Edge
22%Real-time DETR optimized for transformer-friendly NPUs. Wins on transformer-class silicon, struggles on legacy NPUs.
Object detectionApache-2.0INT8FP16transformercocodetection
92K downloads 5.4K deploymentsUpdated Apr 2, 2028
Headline:18.6ms · NVIDIA Jetson Orin Nano · FP16
HIL simulation runs
End-to-end runs of RT-DETR-Edge in scenario rigs that match real customer conditions. Each tile is an actual playable HIL session — sustained throughput, in-sim accuracy, intervention counts, the whole trace.
48 FPS94.1% acc
Open
AV · Urban
CARLA urban · night fog · pedestrians
RT-DETR-Edge×NNVIDIA Jetson Orin Nano
37 min runApr 19, 2028
48 FPS94.1% acc
Open
AV · Highway
Av highway · RT-DETR-Edge
RT-DETR-Edge×NNVIDIA Jetson Orin Nano
37 min runApr 2, 2028
24 FPS88.6% acc
Open
AV · Urban
Av urban · RT-DETR-Edge
RT-DETR-Edge×NNVIDIA Jetson AGX Orin
70 min runApr 2, 2028
18 FPS90.4% acc
Open
Robotics · AMR
Warehouse amr · RT-DETR-Edge
RT-DETR-Edge×NNVIDIA Jetson Thor
61 min runApr 2, 2028
Pro tier opens up scenario authoring
Pro and Team customers can configure custom HIL scenarios — chain CARLA towns, drop in your own actor policies, and pin the exact downstream consumers your fleet runs. Enterprise extends this to dedicated rigs in the lab and on-prem replication.