`M7-M20 Navigation Simulation Environment`

Goal

Extend the existing MuJoCo setup to provide repeatable simulation validation before M20 real-robot tests of planning, behavior, and control changes.

Current Baseline

• The Go1 MuJoCo world and DimOS simulation blueprint are already available.
• Click-to-go, goal publication, logging, and Rerun visualization are available.
• The current setup is the baseline for this work; no new simulator or test world is required for the initial tasks.

Development Flow

Algorithm or configuration change
  -> MuJoCo static-scenario test
  -> MuJoCo dynamic-scenario test
  -> log, Rerun, and metric review
  -> controlled M20 real-robot test
  -> M20 real-world validation

Next Work

• Integrate the M20 model assets and align the simulated sensors
  Load the available M20 model assets into the existing MuJoCo world. Reuse the current Go1/Go2 simulation control and dynamics path initially, align the robot geometry and footprint, and keep the existing sensor interfaces while adjusting sensor position, orientation, FOV, range, and update rate to the available M20 specifications. Full M20 gait and dynamics calibration is not required by this issue.
• Add one moving pedestrian with configurable motion strategies
  Support at least a fixed-waypoint route and a bounded random-route strategy. Use a configurable random seed for repeatable tests, and apply a simple distance-based pause/resume rule around the robot and static objects.
• Add reusable dynamic-obstacle tests
  Add a route-crossing case and a temporary route-blocking case. The simulator must reproduce and record the scenarios; navigation behavior is evaluated by M4-M6 rather than guaranteed by this issue.
• Document the workflow
  Record startup commands, configuration parameters, test procedures, expected simulator behavior, and simulation-platform requirements identified during development.

Required Test Coverage

• Static path generation and tracking smoke test in the existing Go1 test world.
• The fixed-waypoint strategy follows the same configured route across repeated runs.
• The bounded random-route strategy stays inside the configured movement area and is repeatable when the same random seed is used.
• A pedestrian crosses the planned route and the event is visible in Rerun and logs.
• A pedestrian temporarily blocks the route and later clears it.
• The observed stop, no-path, replan, resume, or goal-replacement result is recorded without requiring this issue to change navigation behavior.
• Repeated scenario runs complete without a simulator exception.

Acceptance Criteria

• The available M20 model assets load and run through the existing DimOS MuJoCo blueprint while reusing the current Go1/Go2 simulation control path.
• The sensor and robot interfaces already consumed by the simulation navigation stack remain available.
• Sensor position, orientation, FOV, range, and update rate are configured from available M20 specifications, with any assumptions documented.
• At least one pedestrian supports fixed-waypoint and bounded random-route strategies.
• The pedestrian supports deterministic reset through a configured random seed and applies the documented distance-based pause/resume rule.
• Route crossing and temporary route blocking can be reproduced in the existing test world.
• Results are visible through the existing Rerun and logging workflow.
• Scenario runs complete without a new uncaught simulator exception.
• The issue provides the virtual test environment and records observed navigation behavior; it does not require every planning, control, or recovery behavior to pass.
• Real-robot testing remains the final acceptance source for M20 behavior.

Deliverables

• M20 model-asset integration using the existing Go1/Go2 simulation control and dynamics path, with full M20 gait and dynamics calibration explicitly excluded.
• M20-oriented sensor configuration covering interface compatibility, position, orientation, FOV, range, update rate, and documented assumptions.
• One reusable moving-pedestrian actor with fixed-waypoint and seeded bounded random-route strategies.
• One route-crossing case and one temporary route-blocking case.
• Startup and configuration instructions.
• Test procedure, observed-result notes, and representative Rerun/log evidence.
• A short list of additional simulation-platform requirements found during implementation and testing.

Platform Requirements Identified

The initial work provides concrete input for the continued development of the simulation toolchain:

• Configurable indoor and outdoor scene assets.
• Reusable static objects and dynamic participants.
• Collision, sensor, and robot models with controllable fidelity.
• Scenario reset, parameter variation, and deterministic replay.
• Ground truth, Rerun integration, and quantitative regression.
• Repeatable comparison across algorithm and configuration revisions.
• A consistent workflow from simulation issue reproduction to real-robot verification.
• Support for broader navigation scenarios and future model-training tests.

Future Considerations

A high-fidelity scene-reconstruction and simulation toolchain can support more realistic R&D testing, broader scenario coverage, automated regression, and potentially model training.

A customer-site reconstruction workflow based on ordinary phone capture could also become part of a broader commercial deployment toolchain. It can support pre-deployment validation, expose integration problems earlier, and reduce onsite debugging effort. The commercial workflow has additional product requirements, but it may share scene-reconstruction and simulation technology with the R&D toolchain.

Synced from DIM-1270 by weiyu.liang

Start by locating the existing Go1/Go2 MuJoCo world, DimOS simulation blueprint, sensor interfaces, and Rerun/logging workflow, then run the current static path-generation smoke test. Add the M20 assets and reusable pedestrian scenarios with fixed-waypoint and seeded bounded-random routes, including crossing and temporary blocking cases. Done means repeatable runs record the configured observations and results without new simulator exceptions, with startup and configuration instructions documented.