Conversation
GO2Zenoh's ports, command verbs, mount tree and camera intrinsics move to Go2Base so a second robot-side connection (DDS) shares them. GO2Zenoh keeps its Point-LIO extras: pointlio_map, the odom -> mid360_link edge, the mount tree rooted at mid360_link and the L1 off at start.
Lifted out of go2web's dimos-helper: the CycloneDDS runtime and Go2 types, the sport verb table, and the videohub RTP -> H.264 annex-B reassembly. The module terminates cmd_vel as sport Move (clamped, dead-man StopMove) and command verbs as sport requests, and publishes rt/utlidar/robot_odom with its tf edge, the deskewed L1 cloud, and the front camera. cyclonedds-sys is vendored from our fork with bindgen bumped: 0.55 cannot parse the anonymous unions in CycloneDDS's headers with the libclang the dev shell ships.
Same profile as GO2Zenoh (both are Go2Base), with the rust module as the robot side. The config parity test pins what crosses to the rust struct.
Build goes through a tiny crate flake (nix/, so the path: snapshot never carries target/) with nixpkgs' bindgenHook; the root dev shell change and build.sh go. The binary carries an rpath to libddsc so it runs outside the shell. video_encoding=jpeg polls the videohub GetImageSample at video_fps onto a new image port. rt/lowstate becomes joint_state, imu and battery (decimated to lowstate_hz, battery at 1 Hz), rt/wirelesscontroller becomes joy, and lidar_raw_on adds the undeskewed cloud plus the L1's own IMU. cmd_vel is no longer clamped: limits live in the planner and follower. The command port grows a "<verb> <arg>" form: obstacle-avoidance, joystick, rage, brightness, led, volume, next to the sport verbs and the L1 switch. Go2Base exposes them as rpcs named after GO2Connection's; GO2Zenoh warns on the ones the go2web bridge does not resolve.
The Point-LIO motion stack over DDS: GO2DDS's native process listens on 7447 (the Go2 forwards it to the Jetson, so the viewer still dials go22) and the other processes dial it on loopback. tf_root on Go2Base picks which frame the live odometry moves (GO2Zenoh's mid360_link override becomes its default), odom_tf on GO2DDS drops its own odom -> base_link edge when Point-LIO owns odom.
The robot has no GitHub access, so the private fork comes along like cyclonedds-sys already did. odom_topic and lidar_topic never vary on a Go2 and go; odom_tf folds into tf_root, which now crosses to rust.
The blueprint config layer re-validates every module kwarg from its fields, and the field-less SessionConfig base rejected a pinned router's mode, listen and connect as extras. The ping-pong example's --topology never parsed under `dimos run`; now it does.
The native process refuses a topic it never declared, so a port the python wrapper owns (GO2DDS's camera_info intrinsics) is now named in the config and kept out of the topics handed over.
Point-LIO reads imu from the MID-360; GO2DDS's body IMU on the same topic interleaved a second device into it. Remapped to body_imu.
emit_fine at fine_divisor=2, the planner's own field pitch (FINE = VOXEL / 2 in local_planner/search/se2.py); local_map_fine feeds LocalPlannerNative, the MLS planner and the viewer keep the 0.08 m local_map.
One setting, the local map's voxel: the fine field is half of it, the lattice cell three fine samples, the period two cells, the pad three periods, the snap a fine diagonal. The stock 0.08 reproduces the old constants exactly; both twins, the pyo3 binding and the native config take it. The fine motion stacks plan at 0.04 on local_map_fine.
leshy
changed the title
The Command Center on 7779 came along even with visualization off; a robot-side stack nobody browses now serves nothing.
The fitted sweep's 0° and 180° rows carried 0.416; the feet never leave the trunk walking straight. Drift rows keep the fitted widths.
The 5 cm floor plus the 0.33 body refused a measured 51 cm passage on the accumulated map. test_governor_creeps_in_tight_room still assumes the old floor; untouched pending the real-world run.
This reverts commit 58e6b15.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #4235 (
feat/ivan/go2_dds), which sits on #4203.The motion stacks (
go2-zenoh-motion,go2-zenoh-motion-pointlio,go2-dds-motion-pointlio) turn on the raycaster's fine layer and plan on it at its own pitch.Raycaster:
fine_divisor=2(0.04 m cells),emit_fine=Truepublisheslocal_map_fineon the local-map cadence;LocalPlannerNative.local_mapis remapped onto it. The MLS planner and the viewer keep the 0.08 mlocal_map.Local planner: one setting,
pointcloud_resolution(the map's voxel, default 0.08), replaces the compile-time pitches in both twins. Everything scales off it the way the old constants related: fine field = voxel / 2, lattice cell = 3 fine, period = 2 cells = 3 voxels, pad = 3 periods, snap = fine diagonal (Geom::ofinplanner.rs,pitches()inse2.py). At 0.08 the numbers are the old ones exactly; the pyo3plan()and the native config take it too. The fine stacks passvoxel_size / 2, so they search a 0.06 m lattice on a 0.02 m field.Verified: crate 59 + 5 tests and clippy
-D warnings; both twins plan around the ring fixture at 0.08 and 0.04 and land on the goal; mypy + pytest-fast green. Not yet run on the robot: replan time at 0.04 is 4x the lattice nodes, worth a look on the Jetson.🤖 Generated with Claude Code