You bring the setup.
- The model and controller configuration you use today.
- Any existing logs, in the format you already use.
- An engineer who knows the hardware and can run any agreed tests.
Your policy trusts your sim.
Does your robot agree?
Robot dynamics calibration for sim-to-real teams.
We fit your robot’s simulated dynamics to hardware logs, then test the model on runs it hasn’t seen.
For robotics teams training policies in simulation. Now seeking our first pilot partners.
Start with the motion that doesn’t match.
Your policy works in sim, but on hardware a joint lags, a reversal sticks, or a trajectory drifts. Before another training cycle, check how closely your model predicts the robot’s motion.
We replay hardware commands in simulation and fit dynamics such as friction, damping, and actuator delay. You get a measured comparison of the original and fitted models, so your team can decide what to change next.
One robot. One focused Gap Audit.
A Gap Audit is Zagolo’s scoped robot dynamics calibration pilot for one robot and a repeatable joint-motion mismatch. We review your setup, then agree the parameters to investigate, robot time, delivery schedule, and fixed fee before work starts.
We handle the calibration.
- Log review and a recording plan, including a script if new runs are needed.
- Parameter fitting and an updated simulation model.
- Evaluation on reserved runs and a review of the findings.
What comes back to your team.
A calibrated robot model
An updated model for your agreed simulator workflow, with the fitted dynamics parameters and a record of what changed.
A before-and-after comparison
Trajectory plots and per-joint errors for both models on the same held-out runs. Improvements, regressions, and remaining mismatch are all included.
A clear next step
An assessment of the remaining error, what the recordings can identify, and where the fitted dynamics may help policy transfer. You review the findings with the engineer doing the work.
Judge the model on motion it hasn’t seen.
Set the baseline.
We measure your current model against hardware motion and agree how the comparison will be scored before fitting.
Reserve the test runs.
We set aside complete runs that won’t be used for fitting. Your team keeps the hardware configuration consistent across the comparison.
Freeze the model. Score once.
After fitting, we compare both models on the reserved runs using the same protocol and report every final test result.
Start with what you have.
Tell us your robot, simulator, and the motion that doesn’t match. You don’t need to prepare a data package before we talk.
Have logs? Start there.
We start by reviewing a sample in your existing format. No logs yet? We’ll work out what to capture with your engineer.
We prepare the plan.
If new runs are needed, we prepare the recording script and walk your engineer through the signals and operating limits.
You run it. We take it from there.
Your engineer operates the robot. We handle data review, parameter fitting, and the comparison on reserved test runs.
For your engineer: recording details (optional)
We review commands, joint response, timestamps, and robot configuration together. We agree what to capture before any new runs and help plan the calibration and test split. Keep existing logs as they are for the initial review.
Read the robot recording guide Download the technical referenceA few things you may be wondering.
Which robots and simulators are a fit?
Start with one robot, an existing model, and a repeatable joint-motion mismatch. We review your embodiment and controller, then confirm whether an Isaac, MuJoCo, or Genesis workflow fits the pilot.
Can we start with logs we already have?
Yes. Start with a sample in the format you already use. We review the signals and timing, then explain what is usable and whether any new runs are needed. No need to reformat your logs before we talk.
What improvement do you promise?
We do not promise a preset reduction. These first pilots will establish what is achievable on real hardware. The report includes improvements, regressions, and unresolved error.
Will a better fit make our policy work?
A closer dynamics fit may help, but it does not establish task success. Perception, contacts, and task design can still limit transfer. The audit evaluates motion prediction; a policy test on hardware is a separate experiment.
Why use Zagolo if we can calibrate in-house?
Your team supplies the hardware context; we take on the fitting and evaluation work. The pilot gives you a dedicated engineer, an updated model, and a measured comparison while your team continues its policy work.
What are the timing and terms?
Robot time, delivery schedule, scope, and a fixed fee are agreed in writing after we review your setup.
How will you handle our robot logs?
Robot logs are used only for calibration. We do not store them unless a signed written agreement explicitly covers storage. Before transferring confidential logs, we agree the transfer method, access, retention, and deletion arrangements.
Do we have to publish our data?
No. A public case study, use of your team’s name, and release of recordings or models require a separate agreement with you.
Let’s look at the motion that doesn’t match.
Tell us your robot, simulator, and what moves differently on hardware. We’ll talk through your setup and work out the next step together.
No data package needed. A quick description is enough to start.
