Founding partner programme open for robotics teams
It failed in the field. Make it fail on the bench — the same way, every run.
Some perception failures start below the log: a camera link that drops, a frame that arrives late, a timestamp that slips. Logicatoms replays your robot's own recorded data with those sensor faults injected in hardware, so a failure you saw once becomes one you can reproduce, debug and fix.
Replaying the log replays the pixels, not the moment the sensor path broke.
Simulation and software fault injection can model what a sensor fault does to the data. They can't make your real drivers, image pipeline and clocks react to it. Many of the failures that cost the most days live in that reaction: how the system recovers when the link drops, not the missing frame itself.
-
Link drops and recoveries
Heat, vibration, a loose connector or EMI break the link for a few frames. The log shows a gap, or nothing at all.
-
Late, repeated or reordered frames
The frame the stack acted on was not the frame it thought it was. Replay delivers frames on time, every time.
-
Timestamp skew between sensors
Fusion quietly goes wrong when camera and IMU disagree about time by a few milliseconds.
-
Rare, so it stays unfixed
If it happens once a week on one robot, it can take weeks to see it again. Nobody can fix what they can't reproduce.
Your data. Real faults. The same result every run.
Start from your recorded run
Your own log from the field, streamed into your perception compute exactly as the sensors would deliver it.
Inject the fault in hardware
Faults modelled on real field failures are injected on the sensor path itself, in FPGA, at a precise frame and sample, so your real hardware and drivers have to recover from them.
Get the same result every time
Every run is bit-identical, so you can see exactly what your stack did, change one thing, and run it again.
What you get
- A replay reportYour log, the fault injected, and what your stack did, frame by frame.
- A fault catalogueSensor failures characterised from real field experience, growing with every engagement.
- The same faults in trainingA software version of each fault for data augmentation, matched to the hardware so you test against what you trained on.
- Hardware for your own benchThe injection and replay cores, licensed to run on your validation setup.
Built for the people who own the failure.
Perception and validation leads
At warehouse robots, AMRs, drones and autonomous machines, closing field issues before the next release.
Camera and compute-module teams
Applications engineers fielding customer escalations that end in "could not reproduce".
HIL and system-test engineers
Adding sensor-level fault coverage to an existing validation bench.
Founding partners, now.
We're looking for a small number of robotics teams to build the first fault catalogue with, from real field failures. If your team has a perception failure you couldn't reproduce, that's exactly where we want to start.
- A 30-minute conversation about your last few field failures
- We pick one, together, that your current tools couldn't reproduce
- We replay it on the bench and show you the result
Chip-design discipline, applied where perception meets the sensor.
Logicatoms was founded by a silicon engineer with more than twenty years of chip design at AMD, Intel, Marvell and EdgeQ, from architecture and RTL through verification and post-silicon bring-up.
In chip design, a failure you can't reproduce is a failure you can't fix. We bring that discipline to robot perception: deterministic hardware, exact timing, and evidence your team can check for itself.
Tell us about a failure.
What happened, what sensors were involved, and why it was hard to reproduce. A few lines is enough to start.
info@logicatoms.com
Bangalore, India