Esacode

Solutions · Autonomous vehicles

Edge cases on demand for perception and planning

The rare events are the ones that matter, and they are the hardest to collect. EsaSim runs large scenario sweeps on your own GPU fleet, so rare events are met in simulation first and anything seen on the road becomes a permanent test.

What EsaSim answers

The questions EsaSim answers for autonomy teams

Rare events decide whether a vehicle is ready. These are the questions a scenario sweep on your own GPU fleet answers before the next drive.

01

How does perception hold up in rain, glare, night and fog?

Replay the same scene under weather and lighting sweeps with sensor noise models, and get detection and tracking metrics per condition instead of one average.

02

Did the new model regress on the scenarios that broke the last one?

Every incident becomes a scenario in the regression set. Each new model runs against all of them, and the results show what got better and what got worse.

03

Which edge cases are worth real miles?

A sweep ranks conditions by failure rate, so the track time goes to the few cases simulation says are risky, not to the ones that already pass.

04

Which of our recorded drives should become scenarios?

Point curation at the drive logs and camera frames where they already live; an AI agent reviews them and keeps the hard cases as a new dataset version, traced to every result that uses it.

Data curation
05

Can we test without drive logs leaving our network?

Workers run on your servers and connect out over HTTPS; logs and recordings stay where they are, and only metrics, previews and the results you choose reach EsaSim.

06

How much testing can our own GPUs do overnight?

Split the servers into compute units and let the queue fill idle nights. Per-member budgets keep one sweep from starving the rest of the team, and usage shows what ran.

Where teams start

Weather and lighting sweeps

Replay a scene under rain, glare, night and fog with sensor noise models.

Perception regression

Run every new model against the scenarios that broke the last one.

Private compute

Keep drive logs on your servers; workers connect out, nothing connects in.

Team quotas

Per-member GPU-hour budgets so one sweep does not starve the rest of the team.

Use the fleet you already paid for

Most AV teams own GPU servers that sit idle at night. Split them into compute units and let the queue fill them.

Usage page: GPU-hours per day over a month, showing overnight sweeps filling idle capacity

Build your first study this afternoon

Create a free account, link GitHub, pair a computer and copy a demo, or describe what you need to your assistant. It takes about ten minutes.