List view
Deployment automation for edge inference. Conversion and optimization: PyTorch to ONNX export module (#80), ONNX to TensorRT optimization for Jetson (#71), conversion pipeline Azure Function (#72), conversion pipeline tests (#75), event-driven conversion trigger on checkpoint save (#77), and ACR push automation (#76). Inference runtime: Dockerfile (#82), inference script for ROS 2 integration (#79), and container build/push integration tests (#81). GitOps delivery: automated manifest update on new model (#84), Flux/ArgoCD configuration for Azure Arc edge clusters (#86), and rollback via Git revert documentation (#88). Split from v0.11.0 to separate deployment automation from security and maintenance.
Due by July 18, 2026•0/12 issues closedEvaluation framework end to end. Delivers the policy rollout engine for offline evaluation (#64), policy inference loop in Isaac Sim (#60), checkpoint selection and A/B comparison (#58), evaluation report export (#56), Isaac Sim Azure VM deployment template (#66), robot URDF loading and environment configuration (#65), unified evaluation metrics aggregation service (#68), CSV export for dashboard data (#67), evaluation dashboard web UI (#69), dashboard integration tests (#70), and the ROVE task-level domain evaluation module (#102). Split from v0.9.0 to separate evaluation delivery from training SDK work.
Due by June 6, 2026•0/11 issues closedHybrid cloud, Azure Arc, and edge fleet management. Caps the v0.x series before GA hardening. Arc and AIO infrastructure: Azure Arc deployment automation script (#83), AIO deployment as Arc extension (#85), Arc + AIO Bicep/Terraform template (#99), and Arc deployment troubleshooting guide (#87). Telemetry and messaging: MQTT publisher library in Python (#94), MQTT topic hierarchy and message schema docs (#97), telemetry dashboard via Azure Monitor (#98), and telemetry integration tests (#100). Edge validation and operations: edge training performance validation (#93), edge-to-cloud connectivity validation suite (#95), air-gapped deployment end-to-end guide (#96), and removal of the obsolete azureml-validation-job-debugging.md doc (#110).
Due by August 1, 2026•4/16 issues closedSecurity hardening, supply chain, and maintenance. Supply chain and signing: SLSA Build Level 3 provenance via slsa-github-generator (#549), Sigstore-signed wheels with provenance for training and evaluation packages (#552), and CycloneDX SBOM emitted alongside SPDX (#554). Security posture: documented and verified branch-protection ruleset (#550), confirmed OpenSSF Best Practices Badge tier with current score published (#551), GitHub Private Vulnerability Reporting verified and documented in SECURITY.md (#553), input validation requirements (#9), and hardening guidance (#10). Maintenance and fixes: src/ directory documentation (#109), broken-link fixes in docs/README.md (#134), SC2318 fix for tf_require local cross-reference (#136), removal of unused instance_type variable in submit-azureml-lerobot-inference.sh (#137), redundant shellcheck disable directive cleanup (#138), JaCoCo parser config removed from codecov.yml (#140), GHSA allowlist documented in dependency-review.yml (#142), setup-dev.sh aligned to shell script conventions (#144), and prerequisite-script error-handling flags (#149). Deployment automation split into v0.13.0.
Due by July 4, 2026•5/22 issues closedEdge recording and upload, plus dataviewer feature work and stability fixes. Edge capture: position-based home-detection trigger (#18), configurable bag chunking by size threshold (#19), chunking and compression configuration docs (#20), GPIO-based episode trigger on Jetson (#22), ROS 2 bag recording node with episode metadata (#24), zstd compression for bag chunks (#25), recording pipeline integration tests (#26), bandwidth throttling configuration (#27), upload queue integration tests (#28), upload telemetry and status reporting (#29), Jetson Orin chunking/compression benchmarks (#30), and resumable upload to Azure Blob Storage (#31). Dataviewer enhancements: multiple camera angles in episode viewer (#436), language/task instruction editing for VLA annotation (#437), VLM-as-judge automated VLA episode scoring (#438), and the incremental frontend framework migration plan for React 19, Tailwind 4, TS 6, and Vite 8 (#462). Stability fixes for hook lifecycle bugs: toast setTimeout cleanup (#473), SSE EventSource cleanup in use-export (#474), and useMemo side-effect removal in use-object-detection (#475). First milestone targeting on-robot software.
Due by June 20, 2026•2/21 issues closedTraining SDK, dataviewer testing, and CI/coverage infrastructure. Ships the training SDK stack: YAML schema for job config (#47), training metrics logging callback (#48), Python SDK for job submission (#49), CLI wrapper (#50), Azure ML compute and environment template (#51), training SDK unit and integration tests (#52), monitoring/troubleshooting docs (#53), auto-retry on failure (#55), and automated checkpoint saving to blob (#57). Resolves Azure ML RL submission path and runtime dependency failures (#320). Expands dataviewer test coverage toward OpenSSF Silver: TypeScript frontend testing infra tracking (#210), custom hooks (#214), API client and data layer (#215), shared utility libraries (#216), dashboard components (#219), episode viewer (#218), annotation panel (#220), frame editor (#221), remaining feature components (#222), and backend Python coverage expansion (#441). Adds platform glue: dataviewer feature-flag system with ONNX detection backend option (#435), broader pytest coverage measurement (#442), OSMO log retrieval fix under workload identity (#491), copilot-setup-steps workflow for cloud coding-agent provisioning (#497), and codecov component groupings for dataviewer backend and frontend (#505). Evaluation work split into v0.10.0.
Due by May 23, 2026•2/27 issues closedCloud data pipeline and platform foundations. Delivers ROS bag to LeRobot parquet conversion (#38), Event Grid blob-upload triggers (#32), Azure Fabric pipeline orchestration (#34), frame-gap frequency analysis (#33), validation report schema and generator (#35), synthetic-data validation test suite (#36), conversion architecture docs (#37), conversion pipeline IaC template (#39), BYOK encryption (#41), validation integration (#42), multi-tenant RBAC (#43), vision-based frame quality validation (#44), optional OneLake shortcut config (#46), and supporting AML infrastructure improvements (#384). Also includes platform hygiene: tflint developer setup docs (#206), TFLint soft-fail enforcement (#283), Go/Terratest contract and integration tests (#290), Go added to CodeQL matrix (#335), dev-container tool fix for postgres/redis purge cleanup (#500), and stale binary dependency updates (#522). Foundational data plane plus platform readiness for downstream training and evaluation.
Due by May 9, 2026•2/22 issues closed