[GSOC 2026] Project #8: Local Obstacle Avoidance for DORA_NAV

[NaveenNair25]

Hello dora-rs team,

I have been exploring the Local Obstacle Avoidance project idea and the DORA_NAV repository to figure out the best way to implement this, and having experience working with Nav2's controller, I would like to build a native C++ local planner operator that:

• Subscribes the pointcloud output from mujoco_sim
• Receives local goals from routing_planning
• Outputs direct (v, ω) velocity commands using DWA as the core algorithm

My current assumption is to keep the planner lightweight and reactive, directly producing velocity commands rather than generating immediate trajectories.

Questions:

1. Should the local planner output velocity commands directly to the robot drive, or should it output a corrected short-horizon path that feeds the existing controller?
2. Is DWA the preferred algorithm, or would you like both DWA and APF implemented so users can switch via YAML config?
3. For the simulation demo, should i work within the existing robot_warehouse.xml MuJoCo environment or does the scope include designing a dedicated obstacle course scene?

I look forward to sharing a draft interface sketch and the planned implementation steps alongside the proposal. Looking forward to you clarification on the above.

[chz2012]

Hello,

In general, most navigation architectures generate a segment of path first, and then rely on lateral control and longitudinal control to track the path points. This is because both lateral and longitudinal controllers are tightly coupled with the robot’s chassis and usually need to be adapted specifically for different platforms.

My suggestions are as follows:

• The local planner should output a short segment of path for the controller to track, rather than directly sending velocity commands.
• Following the principle of single responsibility, you may implement two separate nodes (for example, DWA and APF), and allow users to switch between different navigation strategies through dataflow configuration.
• The existing robot_warehouse.xml environment is relatively simple, but it is sufficient for testing. If more complex scenarios are needed, you can load additional scenes or obtain them from the community.

Looking forward to your implementation draft.

[RuPingCen]

Hello,

Thank you for your professional and valuable suggestions, which are very instructive for the overall design of the navigation architecture.

In fact, this planner is specially designed for mobile robots. Considering its motion characteristics and application scenarios, it does not need to adopt the separate lateral and longitudinal control mode. For the algorithm development of this planner, we can refer to the classic DWA or TEB planning algorithms in ROS2 for implementation.

Best regards