The same model that anticipates car crashes anticipates conflicts for your robot.
A 22M-parameter collision-anticipation layer, trained on 100M real-world miles a month from 350,000 active dashcams. It transfers zero-shot to machines it has never seen, with no retraining. Drop it onto the machine you’ve already shipped.
.svg)
Drop your footage.
See collision probability per frame.
Upload a clip from any camera that moves through the physical world – dashcam, forklift, sidewalk delivery, drone. BADAS 2.0 outputs a graded probability per frame. Your planner decides what to do with it.
A layer, not a stack. Drop it onto the machine.
BADAS 2.0 outputs a graded collision probability per frame. Your planner decides what to do with it. We don't replace the certified stack – we sit beside it as the long-tail anticipation layer.
Tell us what your machine sees.
Drop a clip and book a 30-minute scoping call. We'll show you the per-frame probability on your environment.
Five things to know.
The headline numbers, in priority order. Per-category benchmark tables below.
Runs zero-shot on platforms it has never seen.
Generalizes across machines, with no retraining.
22M params. <3 ms on A100. <60 ms on Jetson Nano. Real-time on Orin.
A layer, not a stack.
100M real-world miles a month, 350,000 active dashcams. Zero synthetic.
AUC and AP across 10 scenario groups (888 clips, sliding window). BADAS 2.0 leads in every category. Best per row in bold.
2.0
2.0
2.0
1.0
Open
Reason2
Pro
VL-2B
Single-window mean AP over three lead-time thresholds (1,344 clips). BADAS 2.0 improves mAP from 0.925 to 0.940 while cutting the false positive rate by 74%.
On the internal test set, BADAS 2.0 cuts the false positive rate from 17.7% (v1.0) to 4.6% – a 74% reduction with no loss of recall.
Fraction of collision events detected before they occur (888 clips, 10 scenario groups, threshold 0.75).
AUC and AP on three public academic benchmarks using ego-centric re-annotation. Best per column in bold.
Average collision probability over normalized pre-event time (0% = start, 100% = event). Each clip's timeline is scaled independently, so clips of different lengths are comparable. Positive clips only. BADAS models ramp up sharply; competitors stay flat.
.png)
.png)
.png)
.png)
.png)
Reading this chart: For every positive clip, each model's prediction timeline is normalized so 0% = first prediction and 100% = labeled event. Predictions are binned into 10 equal intervals then averaged across clips. Curves are baseline-normalized per model so the y-axis shows each model's rise above its own floor. A steep ramp means confidence increases sharply as the event approaches; a flat line means the model outputs a near-constant score regardless of proximity to collision.
A graded probability the planner can act on.
BADAS 2.0 outputs an attention heatmap and a per-frame collision probability. Drop both into your planner, your fleet console, or your audit log. The model is a layer; the planner decides.
Runs on the silicon you've already shipped.
Three sizes of the same model, validated on the silicon robotics teams actually deploy on – A100 in the cloud, Jetson Thor / Orin in the rack, Jetson Nano on the edge. Same architecture, same training, same world model.
- SOTA performance across all metrics
- Best mTTA and early warning recall
- Expert in rare long-tail cases
- 34ms on A100 · 41ms on Jetson Thor
- End-device-optimized model
- Expert in false alarm prevention
- Outperforms BADAS 1.0 on every metric
- 4.8ms on A100 · 12.5ms on Jetson Thor
- Ultra-light model for IoT devices
- Optimized for GPU and CPU inference
- Rivals BADAS 1.0 at 14x fewer params
- 2.8ms on A100 · 5.9ms on Jetson Thor
0.994 AP vs 0.940 on the same data.
We trained NVIDIA's COSMOS-Reason2-2B on the exact same 2M clips BADAS 2.0 was trained on, and published the result. BADAS 2.0 lands 99.4% AP at 91x fewer parameters.
COSMOS-BADAS = NVIDIA COSMOS-Reason2-2B fine-tuned on the same 2M Nexar training clips used by BADAS 2.0.
How it actually works.
BADAS 2.0 fine-tunes V-JEPA2, the architecture Yann LeCun proposed for world models. The point isn't the lineage. The point is that latent-space prediction beats pixel reconstruction on collision anticipation, and we publish the benchmarks to prove it.
BADAS 2.0 fine-tunes a V-JEPA2 ViT-L backbone (300M parameters, 24 transformer layers) end-to-end on 16-frame clips at 256×256 resolution and 8 fps. A future-prediction branch estimates the scene 1 second ahead and concatenates it with the current clip, giving the prediction head access to both present evidence and near-future dynamics. Domain-specific SSL pre-training on 2.25M unlabeled Nexar edge device clips is the critical enabler for the distilled edge variants.
~200,000 Labeled Videos. Zero Synthetic Data.
Most collision anticipation models are trained on synthetic data or small academic datasets. BADAS 2.0 is trained exclusively on real-world edge device footage from Nexar's network – the largest ego-centric driving dataset ever assembled for this task.
BADAS 2.0 is trained on ~200,000 labeled videos (~2M windowed clips) – a 5x expansion over v1.0. The corpus is assembled through intelligent data mining: BADAS 1.0 runs as an active oracle over millions of unlabeled Nexar drives, surfacing high-risk clips for human review.
The result: 99.4% AP at 4.6% FPR – a 58% reduction in false alarms over v1.0 on the sliding-window benchmark, with gains across all subgroups including the hardest long-tail categories.
.svg)
.svg)
.svg)
.svg)
Frequently Asked Questions
22M parameters. 100M real-world miles a month. Drop it onto your machine.
Upload a clip and see what BADAS 2.0 predicts on your environment. Or book a 30-minute scoping call – we'll walk through how it lands on the silicon you've already shipped.
Get In Touch
