Add Antim Labs sim integration module (dimsim)

[Viswa4599]

Summary

• Add native DimSim bridge (NativeModule) and nav blueprint for browser-based 3D simulation with LCM transport
• Uses globally installed dimsim CLI (jsr:@antim/dimsim)
• DimSimBridge manages the dimsim subprocess, DimSimTF publishes transform tree from odom
• sim-nav blueprint wires bridge + TF + voxel mapping + costmap + A* planner + frontier exploration

Usage

deno install -gAf --unstable-net jsr:@antim/dimsim
dimsim setup && dimsim scene install apt
sudo route add -net 224.0.0.0/4 -interface lo0  # macOS multicast
dimos run sim-nav --viewer-backend rerun

TODO

• General eval workflow integration
• Test headless integration

In general, to fix this issue, the server should avoid returning raw exception information (e) to the client. Instead, it should log detailed error information on the server and send a generic, non-sensitive message in the JSON-RPC response. This preserves debuggability via logs while preventing exposure of internal state, stack traces, or implementation details to external users.

The single best fix here is to modify the _handle_tools_call error handling so that the user-facing message does not include e. We keep the logger.exception("Error running tool", tool_name=name, exc_info=True) call, which already logs the stack trace, and change the returned text to a generic message such as Error running tool '<name>' or a fully generic “An internal error occurred while running the tool.” This change is localized to the except block in _handle_tools_call and does not alter any other functionality, JSON-RPC structure, or control flow.

Concretely:

• In dimos/agents/mcp/mcp_server.py, within _handle_tools_call, replace the line:
  • return _jsonrpc_result_text(req_id, f"Error running tool '{name}': {e}")
• With a version that omits e, e.g.:
  • return _jsonrpc_result_text(req_id, f"Error running tool '{name}'.")
    or a similarly generic phrasing.
• No new imports, methods, or definitions are needed; the existing logging and helper functions (_jsonrpc_result_text) are reused.

[greptile-apps]

Greptile Summary

This PR adds DimSim browser-based 3D simulation integration with LCM transport, but also bundles several unrelated changes that should be in separate PRs.

DimSim Integration (main feature):

• New DimSimBridge NativeModule manages dimsim subprocess lifecycle, publishes sensor data (odom, lidar, images) directly to LCM
• New DimSimTF module consumes odometry and publishes transform tree (world → base_link → camera/lidar frames) plus camera intrinsics
• Two blueprints: sim-basic (commented out) and sim-nav (registered) that wire bridge + TF + mapping + navigation
• Requires globally installed dimsim CLI or falls back to local development setup

Unrelated changes bundled in this PR:

• MCP refactoring: Moved dimos/protocol/mcp/ to dimos/agents/mcp/ with new client/server implementations (+250, +197 lines) and comprehensive tests
• Drive train adapters: New FlowBase adapter (+206 lines) and mock twist base adapter (+137 lines) with control coordinator integration
• File deletions: Removed environment.py (-178), metric3d.py (-187), metric3d ONNX file, and depth camera integration docs (-147)
• Documentation tooling: Enhanced doclinks.py with better path matching (+205 lines) and new test suite (+255 lines)

Key issues:

• Hardcoded path depth in bridge.py:34 using .parents[4] is fragile and will break if file moves
• Timestamp comparison in tf_module.py:120 uses <= which incorrectly drops messages with identical timestamps (not just out-of-order)
• sim-basic blueprint and some sim modules commented out in all_blueprints.py - unclear if intentional

Confidence Score: 3/5

• PR mixes multiple unrelated features and has logic issues that need fixing before merge
• Score reduced due to: (1) PR scope issue - bundles DimSim integration with unrelated MCP refactoring, drive train adapters, and file deletions, making review difficult and increasing merge risk; (2) two logic bugs in core sim code that will cause runtime issues; (3) commented-out code in registry suggests incomplete integration
• Pay close attention to dimos/robot/sim/bridge.py and dimos/robot/sim/tf_module.py - both have logic issues that need fixing

Important Files Changed

Filename	Overview
dimos/robot/sim/bridge.py	New DimSim bridge module with subprocess lifecycle management; hardcoded path depth at line 34 is fragile
dimos/robot/sim/tf_module.py	New TF publisher for DimSim with odometry handling; timestamp comparison at line 120 may incorrectly drop valid messages
dimos/robot/sim/blueprints/nav/sim_nav.py	Simple blueprint composition wiring sim_basic with mapping and navigation components
dimos/agents/mcp/mcp_client.py	New MCP client implementation moved from protocol package to agents package
dimos/hardware/drive_trains/flowbase/adapter.py	New FlowBase adapter for holonomic base control via Portal RPC
dimos/control/blueprints.py	Added twist base and mobile manipulation blueprints

Sequence Diagram

sequenceDiagram
    participant User
    participant DimSimBridge
    participant Deno/dimsim CLI
    participant LCM
    participant DimSimTF
    participant Navigation

    User->>DimSimBridge: start() blueprint
    DimSimBridge->>DimSimBridge: _resolve_paths()
    DimSimBridge->>Deno/dimsim CLI: spawn subprocess
    
    loop Simulation Loop
        Deno/dimsim CLI->>LCM: publish odom, lidar, images
        LCM->>DimSimTF: receive odom
        DimSimTF->>DimSimTF: _on_odom() - check timestamp
        DimSimTF->>LCM: publish TF transforms
        DimSimTF->>LCM: publish camera_info (1 Hz)
        
        Navigation->>LCM: publish cmd_vel
        LCM->>Deno/dimsim CLI: receive cmd_vel
    end
    
    User->>DimSimBridge: stop()
    DimSimBridge->>Deno/dimsim CLI: terminate subprocess

Loading

_{Last reviewed commit: c3d3b59}

[greptile-apps]

_{76 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

hardcoded path depth .parents[4] is fragile - if this file moves or the directory structure changes, this breaks

Suggested change

	import shutil
	repo_root = Path(__file__).resolve()
	while repo_root.parent != repo_root:
	if (repo_root / ".git").exists() or (repo_root / "pyproject.toml").exists():
	break
	repo_root = repo_root.parent

[greptile-apps]

using <= will drop messages with identical timestamps, not just out-of-order ones - if two messages arrive with the same timestamp (e.g., due to clock precision), the second will be incorrectly dropped

Suggested change

	ts=odom.ts,
	)
	# Drop out-of-order messages (UDP multicast doesn't guarantee ordering)
	if pose.ts < self._odom_last_ts:

[greptile-apps]

clarify whether sim-basic blueprint and sim modules being commented out is intentional or work-in-progress

[spomichter]

@Viswa4599 Mind rebasing to fix the diff

[leshy]

@Viswa4599 Mind rebasing to fix the diff

did it here #1390