Oriens Systems

Manufacturing data and environments

CAM data for model training.

We develop datasets for CAM generation, pairing CAD geometry, CAM projects, and NC programs with their associated generation traces.

Synthetic samples are currently available. Shop-data collection and reinforcement learning environments are in development.

View samples ↓

Data

CAM training requires paired examples of part geometry, process planning, program generation, and revision. We are developing two complementary data sources: synthetic CAM workflows and records of machining practice collected from manufacturing facilities.

Synthetic data

Available now

Synthetic examples pair input CAD geometry with generated CAM projects, toolpaths, NC code, and model traces. Each package preserves the relationship between the generation process and its resulting artifacts.

The generation workflow incorporates material-removal simulation and collision detection. These records are intended to support training on CAM generation and iterative revision using simulation feedback. All samples presented here are synthetic.

View samples ↓

Real shop data

Coming soon

We are collecting CAM projects and NC programs executed on physical machines, together with records of the machine configuration, material, tooling, and workholding.

Collection also includes machinist decisions and program revisions made to satisfy part requirements. These records provide manufacturing context for studying how CAM strategies are adapted to specific equipment and production constraints.

Targeted capabilities

The intended training objective is end-to-end CAM generation under geometric, machine, and process constraints, with reduced reliance on manual correction. Dataset development targets the following capabilities.

Geometry and feature interpretation

Identification of machinable features and interpretation of geometry, tolerances, and surface requirements to inform operation selection and process planning.

Setup planning and workholding

Selection of part orientations, datums, fixtures, and setup sequences subject to accessibility and workholding constraints.

Tool and parameter selection

Selection of tooling and cutting parameters conditioned on material and geometry, accounting for tool reach, rigidity, and deflection.

Complete toolpath generation

Generation of roughing and finishing sequences, including entry and exit moves, linking motions, clearances, and tool changes.

Machine-specific NC

Post-processing for a specified machine and controller, including work offsets, tool calls, and controller-specific commands.

Simulation and revision

Identification of collisions, gouges, and residual stock, followed by program revision and reassessment using simulation feedback.

Error recovery

Diagnosis of failed operations, invalid toolpaths, and CAM software errors, followed by corrective changes and regeneration of affected operations.

Machining performance

Selection of machining strategies that balance cycle time, tool wear, and surface finish while satisfying dimensional requirements.

Sample data packages

The sample below illustrates the contents of a synthetic data package. Full packages are available on request.

Model traces

Recorded model interactions during CAM generation, paired with the corresponding artifacts.

Request full package ↗

Toolpath simulation

Recorded tool motion and material removal with timeline playback.

Accelerated replay with isometric, top, front and side views. Views follow each setup; playback does not represent machine cycle time.

CAM software integration

Coming soon

Planned integration with CAM software interfaces to support native project creation and modification, including setup creation, operation configuration, toolpath regeneration, and NC export.

RL environments

Coming soon

We are developing reinforcement learning environments for CAM generation and revision, with feedback derived from geometric and machining checks.

The verification objective is to assess conformance to part requirements and machining constraints across alternative toolpath strategies.

Benchmarking

Coming soon

Team

Jaffer Razavi

UWaterloo mechanical engineering. Worked on Bruce’s Unit 3 refurbishment and Lu-177 isotope production system.

At Fuse, worked on pulsed-power fusion hardware, FAETON-I, and pulse generators with <1 ns jitter.

jafferr.com ↗

Anwar Al-Shamkhany

Western software engineering. Worked with the Canadian Space Agency on Artemis III and Canadarm2, developing embedded flight software and robotic control systems.

Also worked on p-bits and probabilistic computing.

LinkedIn ↗

Shakeel Hussein

Entered dental school at 15. Joined the Twitter/X transition team at 19.

Built Ridges, an open-source coding-agent competition that achieved 73.6% on SWE-bench Verified and reached approximately $100M in market capitalization. Acquired by Latent Holdings.

shakeel.com ↗

Interested in the data?

Contact us to discuss dataset access and coverage requirements for part geometries, machines, and materials.

hello@oriens.systems ↗