Industry · 03 / Autonomous vehicles

Reconstruct every mile, frame by frame.

LiDAR, camera, IMU and brake-by-wire on one timeline — replayable against ground truth, from test track to public road.

100 Hz
vehicle bus beside 10 Hz lidar, in sync
Scene-true
every disengagement replayable as recorded
Road → sim
one pipeline from log to simulation
01 · Phloem for autonomy

Every disengagement, explained.

When the scene, the signals and the stack’s decisions live on one playhead, triage stops being archaeology.

Scrub the whole scene

Point cloud, camera, IMU and actuation snap to the same instant — with detections and masks overlaid on the frame.

Replay the road into sim

Feed the exact recorded scene into simulation and run the new stack against it before it drives again.

Mine the fleet for edge cases

Query every mile for the same scenario signature and turn one incident into a coverage set.

02 · Lifecycle

One platform, from first prototype to fleet.

The record built in development carries into validation and operations — nothing starts from scratch.

01

Develop

  • Stream track testing and shadow-mode logs into one timeline.
  • Compare stack versions on identical scenes, decision by decision.
  • Keep sim runs and road data in the same record.
02

Validate

  • Run scenario suites against every release candidate.
  • Codify safety criteria once; every mile is scored against them.
  • Build regression libraries from real disengagements.
03

Operate

  • Capture the deployed fleet continuously, within bandwidth budgets.
  • Surface off-nominal behaviour as it happens, fleet-wide.
  • Trend sensor degradation before it becomes a fault.
03 · One timeline

What a run looks like in Phloem.

A shadow-mode drive as it lands: point cloud, forward camera, IMU and the vehicle bus, already on one playhead.

Route 12 · Shadow vehicle · recordingcloud · in sync VEH-07 · LIVE
41:05elapsed
5,412Kmessages
9streams
12.6 GBon disk
T+00:00T+01:00T+02:00T+03:00T+04:00
lidar/lidar/points 10 Hz
video/cam/front 30 fps
imu/imu/accel_z 9.81 m/s²
bus/can/brake_bar 34 bar
The scene, reconstructed

The point cloud, the camera frame, and what the stack believed — detections and masks overlaid — scrubbed together at the moment of disengagement.

vehicle points · 3d ped · 0.87 /cam/front · boxes2d · seg
04 · Ecosystem

Native to the AV stack.

First-class ingest for the formats, protocols and buses this work already runs on — and the wider ecosystem behind them.

frameworkROS 2
formatMCAP
simIsaac Sim
sceneOpenUSD
trainPyTorch
protocolCAN bus

And anything else with an SDK, a topic, or a recorded log.

05 · Outcomes

What teams get back.

Minutes
Per disengagement

Triage opens on the reconstructed scene, not on a directory of logs from three recorders.

1 pipeline
Road to sim

The same recording that explains an incident becomes the scenario that prevents the next one.

06 · End of run

Miles that make the stack smarter.

Every mile driven is training signal. Keep all of it, in sync, in one place.