The Navigation Layer: Kompass Roadmap#

Shipped in 2026

  • Mar 2026 — release notes — Pure Pursuit controller (C++ + Python bindings), joining the existing DWA controller family. Depth Detector (2D→3D via MAD outlier rejection). Tiling for arbitrary-length GPU kernels. Sub-group reductions. Critical Zone Checker direct-lidar GPU kernel. Benchmarking harness with power monitoring.

  • Apr 2026 — release notes — GPU-side pointcloud→occupancy pipeline (~0.57 ms for 100k points on RTX A5000). DWA adaptive prediction horizon. Arc-length goal cost (CPU + GPU parity). RGBDFollower refactor.

  • May 2026 — ROS layer — release notes — Planner accepts Detections / Trackings / PointsOfInterest as goals from multiple topics. Pure Pursuit exposed at the ROS layer. Frame-contract alignment across the stack. Documentation overhaul to developer-style.

  • Aug 2026 — release notes — Zero-copy sensor boundary (pointcloud→grid 90× speedup; emergency-stop up to 587× CPU / 386× GPU). Python 3.14 support. GIL released during solves. NaN hardening.

  • Aug 2026 — release notes — Multi-sensor pointcloud fusion (N clouds → 1 grid, 1 emergency stop; fusion effectively free). Full 3D sensor mounts (position + quaternion, complete [R|t]). All 8 PointField encodings supported. DWA thread-pool reuse.

In Flight — Q4 2026 → Q1 2027

  • Multi-alternative global paths — kompass#44 — planner returns ranked alternative paths so recipes and higher-level agents have real choice.

  • Reactive Station Keeping v2 — enhance the “Intelligent Rest” controller with patterns that use social-force models to stop in crowded human spaces (elevators, lobbies) with minimal movement.

  • Live Self-Healing (nav-specific) — automated recovery for common runtime failures: localization drift, temporary sensor blindness. (Process-level auto-respawn and adaptive horizons are already shipped in EMOS 0.7.0 and kompass-core 0.8.1.)

Q2 2027 → Q4 2027 — Advanced Mission Scheduling & 3D Navigation

Scheduling & 3D

  • Advanced Navigation Scheduling — Queued Navigation Points with integrated scheduling, allowing robots to manage complex multi-stop routes autonomously.

  • 3D Robot State — extending the global robot state to include 3D pose, orientation, and kinematic constraints for humanoids. (Sensor mounts already full-3D as of 0.8.4.)

  • GPU-Voxel Mapping — real-time 3D voxel grids for local mapping and dynamic obstacle avoidance in unstructured environments.

  • Full 3D Navigation Extension — all navigation components handle ground mobile robots in 3D scenes, including native support for terrain height changes.

HW Release & Analytics

  • Tag-Based Following — UWB high-fidelity tag-following controller in Kompass. Ready-to-deploy dual-channel (AoA/PDoA) robot-anchor + moveable-tag combo released as open-source hardware, with simplified calibration and chained following.

  • Diagnostic Analytics — aggregated fleet-level data collection to identify performance bottlenecks in specific hardware-environment combinations. (Developer-facing benchmarking already shipped in kompass-core 0.8.0.)

Compute Expansion (aspirational)

  • Adreno / Mali via OpenCL — test and benchmark kernels on additional integrated GPUs (Qualcomm Adreno, ARM Mali) via an OpenCL target, complementing the existing SYCL / AdaptiveCpp / CUDA path. See current benchmarks.