Industrial Vision: Reliability Uplift in Bin-Picking

Roboception GmbH builds stereo‑based perception solutions for industrial robotics. During my time there, I designed and implemented core software for the flagship products rc_visard (smart stereo sensor) and rc_cube (edge compute platform), and improved grasping strategies within the proprietary component rc_reason.

Focus areas

Core product software: Components and services powering rc_visard and rc_cube.
Grasping strategies: Design and improvements for object detection and grasp planning in rc_reason.

Highlights

Perception pipelines: image processing, stereo geometry, and 3D reconstruction optimized for factory environments.
Interfaces and data flow: tighter coordination between sensing (rc_visard) and computation (rc_cube) to reduce latency and simplify deployment.
Grasp generation and validation: heuristics and scoring to improve successful picks on varied objects and bins.
Production‑readiness: deterministic behavior, robust error handling, and diagnostics for integration teams.

Impact

Faster end‑to‑end cycles in pick‑and‑place setups by tightening the sensing → perception → planning loop.
More stable grasp proposals under challenging lighting and clutter, reducing failed attempts.
Cleaner software boundaries between devices and services for predictable updates and maintenance.

Tech and practices

Languages: C++ (performance‑critical modules), Python (orchestration, tooling).
Domains: stereo vision, pose estimation, point cloud processing, grasp planning.
Practices: testable components, metrics‑driven tuning, CI, containerized builds.

Learnings

Real‑world perception succeeds when algorithms meet solid systems engineering: memory layout, threading, and I/O patterns matter.
Grasping success blends geometry, priors, and pragmatic validation steps tailored to the setup.
Clear contracts between sensing devices and compute services speed up iteration and field support.

References

rc_visard — Smart 3D camera: https://roboception.com/products/rc_visard/
rc_cube — Edge computing platform: https://roboception.com/products/rc_cube/
rc_reason — Perception/grasping components: https://roboception.com/products/rc_reason/