Default track: real UCI robot-failure records with derived RCA and counterfactual analysis. Not a customer production deployment.

ROBOT PICK-AND-PLACE FAILURE INVESTIGATION

Find where a failed robot attempt broke.

REAL-DATA RCA TRACK

UCI Robot Execution Failures

PUBLIC DATA · CC BY 4.0

This track runs on checked-in force/torque traces from the UCI Robot Execution Failures dataset. Telivron derives signal evidence, ranks a bounded hypothesis, and then runs a transparent counterfactual sandbox. It is a real-data engineering demo—not a claim of causal proof.

Open dataset provenance

01 / SELECT A REAL RUN

463 bundled records · 15 samples × 6 channels

Source

lp1.data

Phase

approach

Label

collision

Model

normal

02 / RECONSTRUCT THE EVIDENCE

Fx

peak 69.0

normal-run peak median: 5.7

EVIDENCE SPIKE

Fy

peak 53.0

normal-run peak median: 6.5

EVIDENCE SPIKE

Fz

peak 79.0

normal-run peak median: 56.8

Tx

peak 200.0

normal-run peak median: 15.3

EVIDENCE SPIKE

Ty

peak 109.0

normal-run peak median: 15.4

EVIDENCE SPIKE

Tz

peak 2.0

normal-run peak median: 6.0

SampleFxFyFzTxTyTz
1245379-2001092
2691448-401000
315-145524112
4-9-206219-162
511-8536130
610266-6130
7-8468-10-81
8-12558-10-181
91659-1110
103364-940
11-2171-6-30
122159-630
131263-7-10
14-3-161-4-5-1
15-3-264-3-5-1

Observed evidence: force/torque trace only. The public record does not include camera, policy, planner, firmware, or physical-object context.

03 / RCA HYPOTHESES

MOST SUPPORTED FROM AVAILABLE DATA

collision signature in the force/torque trace

The record is labeled “collision” in the source dataset. A peak magnitude of 79.0 appears in the observed trace and is the strongest available signal for prioritizing this failure class.

66/100 support

NOT ESTABLISHED BY THIS DATA

  • — Whether the cause was controller, hardware, or object interaction
  • — What the camera or policy predicted
  • — Whether the same intervention transfers to a physical robot
ALTERNATIVE

Sensor saturation or calibration drift

ALTERNATIVE

Controller response or timing

ALTERNATIVE

Unobserved contact or object interaction

04 / COUNTERFACTUAL SIMULATION

What should we test to avoid this failure?

LOCAL SANDBOX · DERIVED

This is a transparent intervention model derived from the selected trace. It is not Isaac Sim, a world model, or a physics replay yet. A production adapter would replace this function with a customer-approved simulator while preserving the same evidence and review contract.

Baseline replay

Keep the observed trace unchanged.

Reduce contact impulse

Test a slower approach / lower contact impulse scenario.

Hold controller constant

Separate sensor/contact evidence from controller variation.

05 / ENGINEER REVIEW

Confirm whether the hypothesis matches the physical investigation. This label becomes a regression record for future runs.

This prototype stores review state in the browser only. It does not send data to a production system.