Team Falcon - KFUPM

Quanser ACC 2026 - Virtual Stage Submission

Project Links

• YouTube Demonstration: https://youtu.be/LrqUE5EChg8

---

Project Overview

Our team has developed the perception + Main finite state machine and logic for the qcar the navigate within the given map, it integrates a custom-trained YOLOv4-Tiny object detector with a Finite State Machine (FSM) to navigate a 26-node roadmap, adhering to traffic signage and mission parameters.

---

Core Principles

1. Perception

We combine RGB imagery and Depth maps from the realsense on the qcar.

• Perception: getObjectDepthClass block processes raw YOLOv4 bounding boxes and RealSense depth data.
• Depth Extraction: We extract the detected object to calculate the distance_to_object.

2. Localization and Path Planning

The car navigates a global map consisting of 26 predefined nodes (ID 0-25).

• Mission Logic: The algorithm dynamically plans a path from the Hub (Node 10) to a designated pickupNode, followed by a trip to the dropoffNode, before returning to base.
• Coordinate System: All movement is grounded in the QLabs base frame $[0,0,0]$, ensuring 1:1 parity between the virtual and physical roadmaps.

3. Control Systems

The Main_FSM acts as the car's main logic, varying the speed and behavior based on mission state and perception data.

Perception Logic

Our algorithm prioritizes safety via a hierarchical override system. Even during active path execution, the perception layer can seize control to adhere to traffic laws:

Object	Condition	Action
Stop Sign (ID 3)	Distance $\leq 1.0m$	Full Stop for 3.0s; Latch "Served" status to resume.
Red Light (ID 2)	Distance $1.0m$ to $3.0m$	Full Stop; Wait until light is no longer detected.
Green Light (ID 1)	Any Distance	Proceed; Mission logic maintains control.

---

Technical Implementation

The Perception Block (getObjectDepthClass)

The perception code identifies the most confident detection and validates its physical presence. By ensuring bboxes(i,3) > 1 and checking for finite scores, we filter out "false" detections before they can impact the FSM.

The Mission FSM (Main_FSM)

The state machine utilizes persistent variables to track mission progress (arrived1, arrived8, etc.) and perception timers. It includes a Startup Speed Profile to allow sensors to initialize before entering the high-speed mission states.