Feeding the Cloud: Designing Data-Driven Robotics Systems
summary
Demonstration builds a 3D scene graph stored in Neo4j, routes LLM calls through agent gateway, and records traces and costs with OpenTelemetry, Prometheus, Jaeger and Grafana. A soft drone and a Spot-style planner execute planned trajectories from the scene graph; unified routing tests compare external providers (OpenAI, Anthropic, Bedrock, Gemini) and show per-model cost metrics. A scripted failover evicts upstream on 503s and successfully fails over to local Olama, preserving queries and robot operation under intermittent connectivity while highlighting prompt and credential management gaps.
AI-written summary · about
chapters (11)
- 0:00Data-driven robotics intro
- 1:23Robot architectures evolution
- 3:49Neural networks in robotics
- 6:56Cloud integration challenges
- 8:003D scene graph demo
- 10:49Research to production transition
- 13:21Cloud-native tools for robots
- 17:36Universal routing demo
- 21:06Failover scenario demo
- 25:44Cloud-native benefits
- 26:06Conclusion & Q&A
related talks
43mPatterns for cloud-enabled robots
14mReliability Engineering for Hybrid Robot-Cloud Systems
25mBuilding a PaaS for Robotics with Kubernetes
48mBuilding Robotics Applications at Scale using Open Source from Zero to Hero
28mBuild smart robots and operate them easily with AWS Robotics
18mRobotics: why now?
26mDataEngBytes 2023 - MEL-T2-05 - Paul Brebner
21mBuilding a PaaS for Robotics with Kubernetes