add r1lite bringup test suite: hardware-validated scaffolding, config…

scripts/r1lite_test/README.md

@@ -0,0 +1,62 @@
+# R1 Lite Integration — Bring-up Test Suite (scaffolding)
+
+Robot-side ROS 2 validation for the Galaxea R1 Lite, ported from
+[`scripts/r1pro_test/`](../r1pro_test/) (Mustafa's R1 Pro work). These
+scripts talk **plain ROS 2** to the robot — no dimos — so we can prove the
+robot obeys commands before layering the dimos connection Module on top.
+
+> **Status: SCAFFOLDING.** Every topic name, joint count, domain ID, and
+> pose in [`r1lite_config.py`](r1lite_config.py) is inherited from the R1
+> Pro and marked `TODO(recon)`. Nothing here has run against R1 Lite
+> hardware yet. **Run `test_00_recon.py` first and reconcile the config.**
+
+## Known R1 Lite differences from R1 Pro (verify during recon)
+
+| Aspect | R1 Pro | R1 Lite (expected) |
+|---|---|---|
+| Arm DOF | 7 per arm | **6 per arm** (Galaxea spec) |
+| Upper-body joints | 18 (4+7+7) | likely **16** (4+6+6) — confirm |
+| ROS | ROS 2 Humble | docs ship ROS1 **and** ROS2 tracks — confirm which is installed |
+| Chassis 3-gate problem | yes | unknown — test_03 assumes it may differ |
+| ROS_DOMAIN_ID | 41 | unknown — run `domain_scan.py` |
+
+## Order of operations
+
+```bash
+# 0. Find the robot's domain (robot stack must be running)
+python3 scripts/r1lite_test/domain_scan.py
+
+# 1. Recon: dump topics/nodes/rates/joint-counts; then edit r1lite_config.py
+export ROS_DOMAIN_ID=<from step 0>
+python3 scripts/r1lite_test/test_00_recon.py
+
+# 2. Once config is reconciled, run the graded suite (single DDS session)
+python3 scripts/r1lite_test/run_all_tests.py --skip-chassis --skip-arm  # read-only first
+python3 scripts/r1lite_test/run_all_tests.py                            # full, robot moves
+```
+
+All commands run **inside the dev container** (`ghcr.io/dimensionalos/ros-dev:dev`,
+`--network host`) with `source /opt/ros/humble/setup.bash` and the py3.10
+venv active — same environment as the R1 Pro bring-up.
+
+## Files
+
+| File | Purpose |
+|---|---|
+| `r1lite_config.py` | Single source of truth for topics/DOF/domain — **the file to edit after recon** |
+| `domain_scan.py` | Sweep ROS_DOMAIN_ID 0-101 to find the robot |
+| `test_00_recon.py` | Read-only graph dump: topics, nodes, rates, joint counts |
+| `test_01_topic_discovery.py` | Assert expected topics are visible |
+| `test_02_read_arm_feedback.py` | Read left-arm JointState feedback |
+| `test_03_chassis_command.py` | Drive chassis briefly (robot moves) |
+| `test_04_arm_joint_command.py` | Move one arm joint and return (robot moves) |
+| `test_06_torso_command.py` | Move torso to home pose and back (robot moves) |
+| `run_all_tests.py` | Graded runner, one rclpy session, safety confirms between steps |
+
+## Open questions the recon must close
+
+1. ROS 1 or ROS 2 on this unit? (If ROS 1: apply Galaxea's ROS2 upgrade first.)
+2. Real topic names + message types vs the R1 Pro `/hdas/*` + `/motion_target/*` set.
+3. Arm/torso DOF and safe torso home pose (from the robot's startup scripts).
+4. Does the chassis need a gatekeeper / have the 3-gate problem?
+5. RMW implementation + FastDDS version.

scripts/r1lite_test/RUNBOOK.md

@@ -0,0 +1,115 @@
+# R1 Lite Session Runbook
+
+Everything validated 2026-07-02/03, with exact recreate commands.
+Consoles: **[laptop]** = plain laptop shell · **[robot]** = `ssh r1lite` ·
+**[container]** = shell inside `dimos-dev-r1lite`.
+
+---
+
+## One-time setup (already done — recorded for rebuild-from-scratch)
+
+| What | Command | Status |
+|---|---|---|
+| ghcr access (needs `read:packages`) | `gh auth refresh -h github.com -s read:packages` then `gh auth token \| docker login ghcr.io -u KrishnaH96 --password-stdin` | ✅ |
+| Pull ROS-track dev image (NOT `dev:latest` — that one has no ROS) | `docker pull ghcr.io/dimensionalos/ros-dev:dev` | ✅ |
+| Create container | `docker run -d --name dimos-dev-r1lite --network host -v /home/krishnah/krishnah/dimos:/app -v /tmp/.X11-unix:/tmp/.X11-unix -v $HOME/.Xauthority:/root/.Xauthority:rw -e PYTHONUNBUFFERED=1 -e PYTHONPATH=/app -e DISPLAY=$DISPLAY -it ghcr.io/dimensionalos/ros-dev:dev /bin/bash` | ✅ |
+| Rebuild venv as py3.10 (Humble/rclpy) — in container | `cd /app && ln -sf .envrc.r1pro .envrc && rm -rf .venv && UV_PYTHON=3.10 uv sync --all-extras --no-extra dds --no-extra unitree-dds` | ✅ |
+| SSH alias + key | `~/.ssh/config` Host `r1lite` → `r1lite@192.168.1.85`; `ssh-copy-id r1lite` (password: `1`) | ✅ |
+| Persistent laptop IP (2026-07-03) | `nmcli connection modify lidar +ipv4.addresses 10.42.0.100/24` — "lidar" profile now carries 192.168.1.5 AND 10.42.0.100 on every link-up | ✅ |
+
+## Every-session bring-up
+
+1. **[laptop] Verify cable link** (IP is persistent since 2026-07-03 — no nmcli needed):
+   ```bash
+   ping -c 2 10.42.0.2        # robot onboard PC over the cable
+   ```
+2. **[laptop] Container** (after reboot/suspend):
+   ```bash
+   docker start dimos-dev-r1lite
+   docker exec -it dimos-dev-r1lite bash      # one per terminal needed
+   ```
+3. **[container] Per-shell env** (every new shell):
+   ```bash
+   source .venv/bin/activate && source /opt/ros/humble/setup.bash && export ROS_DOMAIN_ID=2
+   ```
+4. **[robot] Stack** (if `tmux ls` shows nothing):
+   ```bash
+   cd ~/galaxea/install/startup_config/share/startup_config/script
+   ./robot_startup.sh boot ../sessions.d/ATCStandard/R1LITEBody.d
+   # wait ~30s (arms/grippers twitch = HDAS health sign), then:
+   tmux kill-session -t r1lite_teleop     # keep GELLO teleop off the arms
+   ```
+   Shutdown: `./robot_startup.sh kill`
+
+## Validated tests (all ✅, safe to recreate)
+
+**Discovery / read-only:**
+```bash
+# [container] full graph recon (topics, nodes, rates, DOF):
+python scripts/r1lite_test/test_00_recon.py
+
+# [container] formal topic assertion (12 expected topics):
+python scripts/r1lite_test/test_01_topic_discovery.py
+
+# [container] live position of any segment:
+ros2 topic echo /hdas/feedback_arm_left --field position
+# ("A message was lost!!!" spam on 200-488Hz topics = slow CLI subscriber, benign)
+```
+
+**Gripper (first actuation ✅ — 0–100 units, 0=closed, 100=open):**
+```bash
+# [container] shell B monitor:
+ros2 topic echo /hdas/feedback_gripper_left --field position
+# [container] shell A — MUST STREAM (-r 10); one-shot --once is ignored.
+# Motion stops the moment you Ctrl-C (dead-man).
+ros2 topic pub -r 10 /motion_target/target_position_gripper_left sensor_msgs/msg/JointState "{header: {stamp: now}, name: [''], position: [85.0], velocity: [0.0], effort: [0.0]}"
+# reopen: same with position: [101.8]
+```
+
+**Arm wrist-roll (✅ — serial joints, independently commandable):**
+```bash
+# [container] shell B monitor:
+ros2 topic echo /hdas/feedback_arm_left --field position
+# [container] shell A — joints 1-5 held at rest values, joint 6 → 0.2 rad:
+ros2 topic pub -r 10 /motion_target/target_joint_state_arm_left sensor_msgs/msg/JointState "{header: {stamp: now}, name: [''], position: [0.065, 0.0, 0.0, 0.008, -0.001, 0.2], velocity: [0.5, 0.5, 0.5, 0.5, 0.5, 0.5], effort: [0.0]}"
+# return home: same with last value 0.001
+```
+
+## NEVER recreate
+
+- **Torso joint targets** (`target_joint_state_torso`) — parallelogram
+  linkage; single-joint deltas made the robot SHAKE (2026-07-03).
+  Galaxea docs: joint+velocity torso signals conflict, disable_torso=true
+  by default. test_06 is hard-guarded. Torso motion later = task-space
+  `target_speed_torso` only, as a designed experiment.
+- **`findrobot_server.sh` on the robot** — rewrites DDS config to
+  discovery-server mode and reboots; would break our multicast setup.
+
+## Chassis: partially solved, one blocker (see BRINGUP_LOG for full saga)
+
+Validated: chassis node accepts streamed targets in RC **mode 5** (RC ON,
+all 4 switches position 1), ramp reaches full speed with acc_limit +
+brake=false streamed (test_03 does all gates itself). RC mode map:
+all-pos1=5 (software), sw1@2+sw2@3=2 (brake), +sw3@mid=3 (RC manual drive).
+
+UNRESOLVED: wheels refuse to turn (both software AND RC manual) despite
+healthy RC/VCU/battery/e-stop. **Next session, check FIRST: the separate
+wireless E-STOP REMOTE** (R1 Lite ships 3 remotes: torso/chassis/e-stop —
+a pressed or dead e-stop fob = failsafe stop, explains everything).
+Fallbacks: Galaxea "Wheel Motor Zero-Point Calibration" doc,
+product@galaxea-dynamics.com.
+
+⚠️ RULES learned the hard way:
+- Chassis node LATCHES the last target forever (no dead-man!) — ALWAYS
+  end chassis work with a zero-velocity stream.
+- Never power the robot on with e-stop pressed (whole-session inhibit).
+- On-robot `ros2 topic list/echo` may show nothing after `ros2 daemon
+  stop` — daemon cache warming, not an outage; wait or use --no-daemon.
+
+## Not yet done
+
+- Chassis wheel-level unblock (e-stop remote lead above), then test_03 PASS.
+- Torso task-space experiment (target_speed_torso, MPC path) — designed
+  session; joint path is guarded off.
+- Mesh copy for LFS: `scp -r r1lite:~/galaxea/install/mobiman/share/mobiman/urdf/R1_Lite/meshes /tmp/`
+- `R1LiteConnection` module (all facts in `r1lite_config.py`).

scripts/r1lite_test/domain_scan.py

@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+"""Sweep ROS_DOMAIN_ID 0-101 and report domains with foreign participants.
+
+DDS participant discovery is multicast and fast on a wired LAN, so a short
+dwell per domain is enough to spot a robot whose domain we don't know.
+"""
+
+import sys
+import time
+
+import rclpy
+from rclpy.context import Context
+from rclpy.node import Node
+
+
+def scan_domain(domain_id: int, dwell_s: float = 0.7) -> list[str]:
+    ctx = Context()
+    rclpy.init(context=ctx, domain_id=domain_id)
+    try:
+        node = Node(f"domain_scan_{domain_id}", context=ctx)
+        try:
+            time.sleep(dwell_s)
+            names = [
+                f"{ns}/{name}" if ns != "/" else f"/{name}"
+                for name, ns in node.get_node_names_and_namespaces()
+            ]
+            return [n for n in names if f"domain_scan_{domain_id}" not in n]
+        finally:
+            node.destroy_node()
+    finally:
+        rclpy.shutdown(context=ctx)
+
+
+def main() -> None:
+    hits = {}
+    for d in range(0, 102):
+        others = scan_domain(d)
+        if others:
+            hits[d] = others
+            print(f"DOMAIN {d}: {len(others)} foreign node(s): {others[:8]}", flush=True)
+        if d % 20 == 0:
+            print(f"...scanned up to domain {d}", file=sys.stderr, flush=True)
+    if not hits:
+        print("No foreign ROS 2 participants found on any domain 0-101.")
+
+
+if __name__ == "__main__":
+    main()

scripts/r1lite_test/r1lite_config.py

@@ -0,0 +1,85 @@
+"""R1 Lite hardware constants — single source of truth for the test suite.
+
+Reconciled against live hardware 2026-07-02/03 (test_00_recon.py, boot
+profile ATCStandard/R1LITEBody.d, unit hostname "r1lite"). Remaining
+unknowns are marked TODO(recon). Full evidence trail: BRINGUP_LOG.md.
+
+Verified robot identity (/opt/galaxea/body/hardware.json): R1-LITE,
+ARM=A1X, TORSO=T0, ECU=E03, CHASSIS=C1(+CS1), ARM_END=EE0,
+HEAD_CAMERA=HC0 (stereo, no depth topic), WRIST_CAMERA=WC1 (RealSense).
+Onboard computer: x86_64, Ubuntu 22.04.5, ROS 2 Humble, user "r1lite".
+"""
+
+# --- Network (verified) ------------------------------------------------
+ROBOT_ETH_IP = "10.42.0.2"     # enp2s0, factory-assigned; laptop uses 10.42.0.100
+ROBOT_WIFI_IP = "192.168.1.85" # office WiFi (wlo1)
+ROS_DOMAIN_ID = 2              # robot .bashrc: ROS_DOMAIN_ID=02; plain multicast
+
+# --- Kinematics (verified via feedback rates/counts) --------------------
+ARM_DOF = 6        # /hdas/feedback_arm_*: 6 joints @ 200 Hz
+TORSO_DOF = 4      # /hdas/feedback_torso: 4 joints @ 488 Hz
+CHASSIS_FB_DOF = 3 # /hdas/feedback_chassis: 3 joints @ 200 Hz
+GRIPPER_DOF = 1    # /hdas/feedback_gripper_*: 1 joint @ 200 Hz
+# /joint_states aggregate: 25 joints @ 500 Hz (naming TODO below)
+# /hdas/feedback_hand_*: exist but silent (no dexterous hands on this unit)
+
+# TODO(recon): safe torso "home" pose — read it from the robot's
+# start_mobiman_torso_speed_control.sh (R1LITE arg) before running test_06.
+TORSO_HOME_POSE: list[float] = [0.0] * TORSO_DOF
+
+# TODO(recon): joint NAMES for each feedback topic (echo --field name),
+# and the 25-joint /joint_states naming; needed for dimos joint mapping.
+
+# --- Topics (verified live 2026-07-03) ----------------------------------
+FEEDBACK_ARM = "/hdas/feedback_arm_{side}"            # JointState, 200 Hz
+FEEDBACK_TORSO = "/hdas/feedback_torso"               # JointState, 488 Hz
+FEEDBACK_CHASSIS = "/hdas/feedback_chassis"           # JointState, 200 Hz
+FEEDBACK_GRIPPER = "/hdas/feedback_gripper_{side}"    # JointState, 200 Hz
+IMU_CHASSIS = "/hdas/imu_chassis"                     # sensor_msgs/Imu
+IMU_TORSO = "/hdas/imu_torso"                         # sensor_msgs/Imu
+CMD_ARM = "/motion_target/target_joint_state_arm_{side}"   # JointState
+CMD_TORSO = "/motion_target/target_joint_state_torso"      # JointState
+CMD_TORSO_SPEED = "/motion_target/target_speed_torso"      # TwistStamped (new vs R1 Pro)
+CMD_GRIPPER = "/motion_target/target_position_gripper_{side}"  # JointState
+CMD_CHASSIS_SPEED = "/motion_target/target_speed_chassis"  # TwistStamped
+CHASSIS_ACC_LIMIT = "/motion_target/chassis_acc_limit"     # TwistStamped
+BRAKE_MODE = "/motion_target/brake_mode"                   # std_msgs/Bool
+CHASSIS_SPEED_FB = "/motion_control/chassis_speed"         # TwistStamped (Gate-1 analog)
+# TODO(recon): R1 Pro's gatekeeper /cmd_vel does NOT exist here — chassis
+# command path is CMD_CHASSIS_SPEED directly; 3-gate behavior unverified
+# (all three gate suspects exist: /controller topic, brake_mode, acc_limit).
+CMD_VEL_GATEKEEPER = CMD_CHASSIS_SPEED
+
+# Perception (verified): wrist cams = full RealSense stacks; head = stereo
+# RGB pair, NO depth topic; NO chassis cameras; NO lidar topic in this
+# boot profile (livox driver installed but not launched).
+HEAD_LEFT_COMPRESSED = "/hdas/camera_head/left_raw/image_raw_color/compressed"
+HEAD_RIGHT_COMPRESSED = "/hdas/camera_head/right_raw/image_raw_color/compressed"
+WRIST_COLOR_COMPRESSED = "/hdas/camera_wrist_{side}/color/image_raw/compressed"
+WRIST_DEPTH_ALIGNED = "/hdas/camera_wrist_{side}/aligned_depth_to_color/image_raw"
+ROBOT_DESCRIPTION = "/robot_description"  # URDF published on-topic (pubs=2)
+
+# Topics that must exist for test_01 to pass.
+EXPECTED_TOPICS = [
+    FEEDBACK_ARM.format(side="left"),
+    FEEDBACK_ARM.format(side="right"),
+    FEEDBACK_CHASSIS,
+    FEEDBACK_TORSO,
+    FEEDBACK_GRIPPER.format(side="left"),
+    FEEDBACK_GRIPPER.format(side="right"),
+    IMU_CHASSIS,
+    IMU_TORSO,
+    CMD_CHASSIS_SPEED,
+    CMD_ARM.format(side="left"),
+    CMD_ARM.format(side="right"),
+    CMD_TORSO,
+]
+
+# --- Remaining open questions -------------------------------------------
+# 1. Joint names per topic + /joint_states 25-joint composition.
+# 2. Torso safe home pose.
+# 3. Chassis gating: does r1_lite_chassis_control_node need subscriber/
+#    brake/acc-limit unlocks like the R1 Pro's 3 gates? (test_03 will tell.)
+# 4. Head depth: stereo-only? (calib/head_{left,right} suggest onboard
+#    stereo depth is possible but not published.)
+# 5. Lidar: present on unit? (No topic; livox driver installed.)