Release planner

[leshy]

#884
#885
#886

and foxglove dash all together

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[greptile-apps]

Greptile Summary

This PR consolidates three related PRs (#884, #885, #886) and adds Foxglove dashboard updates for the GO2 robot. The changes significantly improve the navigation and planning system through better code organization and intelligent replanning logic.

Key Changes:

• Navigation Refactoring: Extracted controller logic into separate PController and PdController classes with proper abstraction, improving code maintainability and testability
• Intelligent Replanning: Added ReplanLimiter class that prevents excessive replanning in the same area while resetting attempts when the robot moves sufficiently far (2m threshold)
• Performance Optimization: Implemented numba-accelerated kernels for occupancy grid generation, providing significant performance improvements for pointcloud processing
• Better Code Organization: Separated concerns in local planner, extracted safe goal finding logic, and improved debug visualization throttling
• Dependency Updates: Upgraded langchain ecosystem to v1.x and switched to a forked unitree-webrtc-connect repository pending upstream PR merge
• URDF Support: Added GO2 robot URDF definition and utility functions for programmatic URDF generation
• Testing: Added new e2e test for spatial memory navigation and expanded occupancy grid tests

Confidence Score: 4/5

• This PR is safe to merge with minor considerations
• The code quality is excellent with well-structured refactoring, proper abstractions, and comprehensive testing. The changes are well-contained and follow good software engineering practices. Score is 4 instead of 5 due to: (1) dependency on forked repository for unitree-webrtc-connect until upstream PR is merged, and (2) major version bump of langchain dependencies which should be monitored for breaking changes
• Pay attention to pyproject.toml for the forked dependency and langchain v1.x upgrade impact

Important Files Changed

Filename	Overview
dimos/navigation/replanning_a_star/controllers.py	New controller abstraction with P and PD controllers for robot navigation, extracted from local_planner for better separation of concerns
dimos/navigation/replanning_a_star/replan_limiter.py	New class to intelligently limit replanning attempts with position-based reset logic
dimos/navigation/replanning_a_star/local_planner.py	Refactored to use controller abstraction, improved debug visualization throttling, cleaner separation of concerns
dimos/navigation/replanning_a_star/global_planner.py	Integrated ReplanLimiter for smarter replanning logic, extracted safe goal finding to separate method
dimos/mapping/pointclouds/occupancy.py	Added numba-accelerated kernels for significant performance improvement in occupancy grid generation
pyproject.toml	Updated langchain dependencies to v1.x, moved numba/llvmlite from overrides to direct dependencies, switched unitree-webrtc-connect source

Sequence Diagram

sequenceDiagram
    participant User
    participant GlobalPlanner
    participant ReplanLimiter
    participant LocalPlanner
    participant Controller
    participant OccupancyMapper
    
    User->>GlobalPlanner: set_goal(goal)
    GlobalPlanner->>ReplanLimiter: reset()
    GlobalPlanner->>GlobalPlanner: _plan_path()
    GlobalPlanner->>GlobalPlanner: _find_safe_goal(goal)
    GlobalPlanner->>GlobalPlanner: _find_wide_path()
    GlobalPlanner->>LocalPlanner: start_planning(path)
    
    loop Navigation Loop
        LocalPlanner->>LocalPlanner: _compute_control()
        
        alt initial_rotation
            LocalPlanner->>Controller: rotate(yaw_error)
            Controller-->>LocalPlanner: Twist
        else path_following
            LocalPlanner->>Controller: advance(lookahead_point, odom)
            Controller-->>LocalPlanner: Twist
        else final_rotation
            LocalPlanner->>Controller: rotate(yaw_error)
            Controller-->>LocalPlanner: Twist
        end
        
        LocalPlanner->>User: cmd_vel.on_next(twist)
        
        alt Obstacle Detected
            LocalPlanner->>GlobalPlanner: request_replan(StopMessage.OBSTACLE)
            GlobalPlanner->>ReplanLimiter: can_retry(position)
            
            alt Can Retry
                ReplanLimiter-->>GlobalPlanner: True
                GlobalPlanner->>ReplanLimiter: will_retry()
                GlobalPlanner->>GlobalPlanner: _plan_path()
            else Max Attempts
                ReplanLimiter-->>GlobalPlanner: False
                GlobalPlanner->>GlobalPlanner: cancel_goal()
            end
        end
    end
    
    Note over OccupancyMapper: Runs in parallel
    OccupancyMapper->>OccupancyMapper: add_frame(pointcloud)
    OccupancyMapper->>OccupancyMapper: simple_occupancy (numba-accelerated)
    OccupancyMapper->>GlobalPlanner: publish costmap

Loading

[spomichter]

@paul-nechifor This seems like a hacky way to be running async dimos run we need a a pythonic equivalent method to run blueprints that don't require a bash subprocess. Any ideas. Ideally blueprint.run()