Selected feed

Illustrative engagements

These stories describe past project patterns — anonymised or clearly hypothetical — not named client endorsements or guaranteed outcomes. Metrics, where mentioned, reflect specific past contexts and are not promises for your site.

Alberta warehouse · AMR perception

LiDAR feed stutter at peak Wi-Fi

An Alberta warehouse running two AMRs for pallet moves had a perception stack that performed adequately through morning picks. By early afternoon, LiDAR point streams began stuttering — not enough to trigger hard faults, but enough that obstacle avoidance started clipping rack corners on tight aisles. Handheld scanner traffic on the same access points was the culprit; ingest buffers had been sized for average load, not peak.

RoboStream rebuilt the ingest layer with adaptive buffering, added channel-health metrics to the ops dashboard, and ran a two-week pilot where human operators retained teleoperation override. Functional safety review confirmed fallbacks when sync quality dropped below agreed thresholds. Post-pilot, near-miss reports in the affected aisles fell — an illustrative result from that engagement, not a guarantee for other sites.

Client review session examining AMR perception pipeline logs

Calgary manufacturer · Sensor fusion

Second camera at half frame rate

A Calgary manufacturer added a rear-facing camera to a cobot cell after a near-miss with a rolling cart. The new stream ran at fifteen frames per second while the existing front camera held thirty. The fusion node had never been scoped for asymmetric ingest; within days, grasp alignment began missing low-profile tooling.

We re-architected the fusion path with explicit per-stream timing metadata, replayed a week of production telemetry through simulation, and recalibrated extrinsics under motion. Human-in-the-loop verification ran for ten shifts before unattended operation resumed. The engagement illustrates how pipeline scope — not model accuracy alone — often defines perception reliability.

Fleet telemetry dashboard showing multi-camera ingest health

Logistics scale-up · Fleet diagnostics

Cold-storage doors and silent telemetry gaps

A logistics scale-up's fleet diagnostics dashboard showed green across five AMRs until a cold-storage door began cycling more frequently each night. Multipath and temperature swing introduced intermittent LiDAR dropouts that the dashboard averaged away. Autonomous navigation did not fail outright — it hesitated, which operators interpreted as software bugs.

RoboStream instrumented per-sensor drop counters, tied alerts to shift handoff runbooks, and validated sim-to-real replay using captured night-shift telemetry. Fleet orchestration policies were updated so units entering the affected zone required explicit human acknowledgment until channel health stabilized. Outcomes improved for that pilot lane; your floor will present its own edge cases.

Industrial SME · Discovery

Bandwidth budget before hardware purchase

An industrial SME planned to add three additional vision streams to an existing AMR without upgrading network infrastructure. Discovery work modelled bandwidth and latency under concurrent forklift traffic and found the plan would breach budgets within six months of scale-up.

Rather than sell additional hardware, we produced a phased roadmap: edge inference for two streams, central fusion for the third, and a retainer for quarterly threshold review. The client avoided a costly mis-purchase — a planning outcome, not a throughput guarantee.

Disclaimer. Case descriptions are illustrative of RoboStream's robotics-AI engineering work. They are not promises of future performance, cost savings, uptime or safety. Robotic systems require risk assessment, applicable safety standards and ongoing human oversight. We do not work on weapons, defence or covert-surveillance applications.