Robots that learn; Operations that thrive.

From data collection to action — DecisionFacts enables robots to sense, understand, reason, act, and learn.

Demonstration Data & Domain Expertise

Expert-demonstrated episodes, captured through teleop, for the exact tasks your robots need to learn.

Simulation, Policy & Fine Tuning

Benchmark and fine-tune ACT, SmolVLA, GROOT, diffusion, and other policy architectures in sandbox before a single hour on real hardware.

Tailored Autonomous Agents

Purpose-built field policies, deployed to edge compute and improved with every run.

The reasoning roadmap

Every dataset and policy we build is reasoning-ready — it includes data for scene grounding, subtask segmentation, motion-grounded spans, causal/contrastive notes (DPO), failure/recovery pairs, and derived VQA pairs.

Scene Grounding

Object identities and bounding boxes at episode start

Subtask Segmentation

Approach → grasp → align → insert → release

Motion-Grounded Spans

Language tied 1:1 to measured action

Contrastive Notes (DPO)

Why this action, not an alternative

Failure/Recovery Pairs

Tied to the per-episode outcome label

Custom Robot Data Collection — Teleoperation, UMI Data, Annotation, and Verification at Scale

Expert operators, production-grade teleoperation hardware, and your dataset format of choice — from pilot episodes to production-scale campaigns.

Data Collection Methods

We choose the operator interface around task precision, throughput, and embodiment — not around a single favored tool.

Best for precision, contact-rich tasks

Leader-Follower Arms (Teleoperation)

Best for precision bimanual and contact-rich tasks with ALOHA-style manipulation.

Looping comparison of a real robot performing a cup-nesting task alongside its simulated policy counterpart

Real robot vs. simulated policy, side by side — the same cup-nesting task from our sample dataset.

Best for fast, portable data collection

Universal Manipulation Interface (UMI)

A handheld gripper with wrist-mounted cameras captures demonstrations with no robot in the loop — quick to scale across tasks, environments, and operators.

Dual wrist-camera views with tracked 3D trajectories from a real UMI episode — opening a bottle with the right gripper.

Simulation and Training

ACT, SmolVLA, and other VLA policies are trained, rendered across multiple camera angles in simulation, evaluated by a vision-language model, and re-trained — before a single hour on your real hardware.

app.decisionfacts.ai
Robot Imitation Learning Pipeline showing simulation configuration, multi-angle MuJoCo renders, and pipeline state including training, simulation, and VLM evaluation metrics

Every move the arm makes is a decision from this trained policy — ACT and SmolVLA can learn directly from task episodes, while GROOT can be fine-tuned from a foundation model on top of them — validated in simulation before a single hour on real hardware.

The same kind of trained policy, running fully autonomously on real hardware — no leader arm, no human in the loop.

Data Quality Methodology

We check for:

1

Synchronized timestamps

All cameras, joints, and actions aligned to <5 ms tolerance using shared clock sources.

2

Consistent episode structure

Identical observation/action schema, dimensions, data types, and key names.

3

Operator qualification

Operators pass a proficiency test on the specific production task.

4

Task success verification

Every episode is reviewed; failures are flagged and separated on request.

5

Scene reset consistency

Object positions, lighting, and workspace state follow documented reset ranges.

6

Frame drop monitoring

Camera streams are checked and episodes above the allowed loss rate are recollected.

7

Gripper state consistency

Gripper signals are validated against visible contact and commanded actions.

8

Joint limit compliance

Trajectories are checked for limit violations, clipping, and unsafe motion.

9

Metadata completeness

Robot, operator, task, calibration, scene, and timing metadata ship with every episode.

10

Annotation standards

Labels follow a documented ontology with spot checks for consistency.

Any arm
hardware-agnostic policies
Sim-first
validated before hardware
Edge-deployed
on-device inference
Proud member of the NVIDIA Inception Program

Ready to Start Your Data Collection or Training?

Whether you need demonstration data, simulation benchmarks, or a policy ready for the field — we'll be happy to assist you in your journey.

Contact us at info@decisionfacts.io
Explore other solutions: Financial Fraud  ·  Supply Chain