[GSoC 2026] Project #8: Local Obstacle Avoidance for DORA_NAV — DWA Local Planner

[sahilshingate01]

[GSoC 2026] Project #8: Local Obstacle Avoidance for DORA_NAV — DWA Local Planner

---

Applicant: Sahil Shingate
GitHub: @sahilshingate01
Project Duration: 175 hours
Chosen Project: Local Obstacle Avoidance for Dora Robots

---

What I'm Building

DORA_NAV currently has no local planner. Global planners tell the robot where to go — local planners keep it from hitting things on the way there. I'm building a DWA (Dynamic Window Approach) local planner operator in C++ that fits cleanly into dora's dataflow model.

At every control cycle it:

1. Computes the dynamic window — the (v, ω) pairs actually reachable given current velocity and acceleration limits
2. Samples velocity pairs across the window
3. Simulates short trajectory arcs for each pair
4. Scores them on heading, clearance, velocity, and angular penalty — outputs the best collision-free command

The operator takes obstacle data (as (x, y, radius) tuples over Arrow) and a local goal coordinate as inputs, and outputs a (v, ω) velocity command. Swapping sensor sources means replacing the sensor operator and updating dataflow.yml — nothing in the planner changes.

---

Prior Work on dora

Before writing this proposal I built dora-inspect — a structural linter for dora YAML dataflows with cycle detection, orphaned node detection, and a WASM browser interface. Built specifically to understand how the dataflow graph works internally.

I also have a merged PR in the main repo: #1347 — invariant tests for NodeExitStatus classification, merged by phil-opp.

---

Deliverables

• C++ DWA operator wired into dora's I/O system with a proper dataflow.yml
• YAML config covering all tunable parameters (velocities, acceleration limits, scoring weights α/β/γ/δ) — no C++ changes needed to tune
• Python 2D simulation (Pygame) with 3 scenarios: random static obstacles, narrow corridor, dynamic obstacle
• pytest suite covering edge cases: dead-end, goal blocked, zero clearance
• Quick-start guide and operator docs written for someone unfamiliar with dora

---

Questions for Mentors

1. Is there a preferred Arrow schema for obstacle data already in use elsewhere in DORA_NAV, or should I define the (x, y, radius) RecordBatch schema from scratch?
2. Any preference on the C++ operator API — are there existing C++ operator examples I should model the structure on beyond what's in the docs?
3. Is Pygame acceptable for the simulation environment, or is there a preferred visualisation tool for dora demos?

---

Full proposal PDF available on request. Happy to discuss any part of the design.