From 50dbba4cc7fb5d66551dd3a8cf3d5c1d17ffec91 Mon Sep 17 00:00:00 2001 From: Ruixiang Du Date: Tue, 7 Jul 2026 16:45:05 +0800 Subject: [PATCH 1/2] build: re-pin navigation (Viz v2: quickviz migration + MPPI tuner) Pins xmNavigation devel at the Viz v2 merge (nav #61): restructured quickviz (318daf4) with cvdraw retired in favor of image/, plus the interactive MPPI tuner on the viewer shell. TODO updated: Viz v2 closed, MPPI M4 progressed to the CPU backend seam (nav #62 pending). Note: quickviz #30 must be merged with a merge commit (not squash) so the pinned 318daf4 stays reachable from upstream main. --- TODO.md | 4 ++-- components/navigation | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/TODO.md b/TODO.md index a62b249..3feea6c 100644 --- a/TODO.md +++ b/TODO.md @@ -19,8 +19,8 @@ The task tracker for the family. Cross-repo sequencing lives here; each componen - [x] W4: telemetry instrumentation across all areas (nav #59) - [x] W5: xmBase v0.4.0 released; umbrella re-pin + full assembly green (#22); **xmNavigation v0.2.0 released** — the algorithm-centric migration arc is COMPLETE - [x] Umbrella: re-pin xmNavigation; `XMOTION_WITH_NAVIGATION=ON` — full assembly green (#22) -- [ ] Viz v2: migrate nav viz to the restructured quickviz (canvas/viewer/plot/scene; cvdraw retired upstream); interactive tuner; evaluate scene/ 3D for SRB + OccupancyGrid renderable (quickviz #30 stb-guard is a precondition) -- [ ] MPPI M4: CUDA backend for dGPU/Jetson Orin (blocked: CUDA toolkit install + reboot on dev machine) +- [x] Viz v2 COMPLETE (nav #61): quickviz migration (cvdraw → image/) + interactive MPPI tuner (cairo alpha-fan world view, live ESS/cost plots, pause/step + λ/σ sliders); still open: quickviz #30 merge (merge-commit, NOT squash — nav pins 318daf4 from that branch), scene/ 3D for SRB + OccupancyGrid renderable +- [~] MPPI M4: rollout-backend seam + threaded CPU backend done (nav #62: SRB 11.0→5.9 ms @4 workers, bitwise-deterministic); CUDA backend still blocked (toolkit install + reboot on dev machine); CPU follow-up: per-sample counter-seeded noise streams (sampling is the serial share, 4.4 ms @K=2048) - [x] Rename component repos + umbrella submodule paths to function words; register xmTelemetry submodule (PR #12) ## Next arc (after the xmNavigation refactor) diff --git a/components/navigation b/components/navigation index fd1e5e8..e907dc7 160000 --- a/components/navigation +++ b/components/navigation @@ -1 +1 @@ -Subproject commit fd1e5e8dfee72196f2604061ec87986cfb1050dd +Subproject commit e907dc7f2dcf53f1a70d0deea6b04c85b17d86e0 From 87947ecc72cd6373a1753e05a502ea9b1a9c6bf4 Mon Sep 17 00:00:00 2001 From: Ruixiang Du Date: Wed, 8 Jul 2026 08:47:51 +0800 Subject: [PATCH 2/2] build: re-pin navigation at post-Viz-v2 devel (CUDA M4 + control pass) Advances the pin from the Viz v2 merge (e907dc7f) to f2dcc4fe, adding nav #63 (CUDA rollout backends: wheeled + SRB programs, on-device Philox sampling), #64 (fsm replaced by the vendored ctfsm v0.2.0 submodule), #65 (output-stage safety shield), and #66 (models consolidation; boost::odeint out of control/). TODO updated accordingly. --- TODO.md | 3 ++- components/navigation | 2 +- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/TODO.md b/TODO.md index 3feea6c..5ea0889 100644 --- a/TODO.md +++ b/TODO.md @@ -20,7 +20,8 @@ The task tracker for the family. Cross-repo sequencing lives here; each componen - [x] W5: xmBase v0.4.0 released; umbrella re-pin + full assembly green (#22); **xmNavigation v0.2.0 released** — the algorithm-centric migration arc is COMPLETE - [x] Umbrella: re-pin xmNavigation; `XMOTION_WITH_NAVIGATION=ON` — full assembly green (#22) - [x] Viz v2 COMPLETE (nav #61): quickviz migration (cvdraw → image/) + interactive MPPI tuner (cairo alpha-fan world view, live ESS/cost plots, pause/step + λ/σ sliders); still open: quickviz #30 merge (merge-commit, NOT squash — nav pins 318daf4 from that branch), scene/ 3D for SRB + OccupancyGrid renderable -- [~] MPPI M4: rollout-backend seam + threaded CPU backend done (nav #62: SRB 11.0→5.9 ms @4 workers, bitwise-deterministic); CUDA backend still blocked (toolkit install + reboot on dev machine); CPU follow-up: per-sample counter-seeded noise streams (sampling is the serial share, 4.4 ms @K=2048) +- [x] MPPI M4a–c COMPLETE (nav #62/#63): rollout-backend seam + threaded CPU backend; CUDA backends (wheeled + SRB programs over shared raw-span cores, float32) + on-device Philox sampling — wheeled end-to-end 0.14 ms @K=2048 / 9.4 ms @K=131072 on GTX 1660 Ti; remaining M4d: device spline-knot sampling for SRB, Jetson Orin deployment/profiling, robot trial +- [x] control folder pass (nav #64/#65/#66): fsm → vendored ctfsm v0.2.0 submodule; safety shield (envelope + ctfsm fallback ladder + CBF-QP barrier, scenario-driven S1–S6); model zoo consolidated into control/models (BicycleAccelModel + dep-free RK4, boost::odeint out of control/); follow-ups: PID production pass, boost removal from state_lattice + reachability internals, S5 quadruped GRF cone shield - [x] Rename component repos + umbrella submodule paths to function words; register xmTelemetry submodule (PR #12) ## Next arc (after the xmNavigation refactor) diff --git a/components/navigation b/components/navigation index e907dc7..f2dcc4f 160000 --- a/components/navigation +++ b/components/navigation @@ -1 +1 @@ -Subproject commit e907dc7f2dcf53f1a70d0deea6b04c85b17d86e0 +Subproject commit f2dcc4fe4f5a8f1a5a0a74a52a07e699dcce95c1