Skip to content

PROBLEM_STATEMENT.md

Latest commit

 

History

History
167 lines (118 loc) · 6.67 KB

PROBLEM_STATEMENT.md

File metadata and controls

167 lines (118 loc) · 6.67 KB
Raw

Level 1 Demo

Level 1 — Operation SKYE-X: Search and Pursuit

Team Aerial Robotics IITK| Y25 Recruitment Hackathon

1. Context

A rogue adversarial drone — designation SKYE-X — has been detected inside a classified industrial facility. Intel suggests it is equipped with onboard evasion algorithms and is actively trying to break line-of-sight, hide behind obstacles, and escape your tracking radius.

You are the operator of an autonomous interceptor quadcopter. You have no direct control — your drone acts on the instructions of the AI controller you write. Your sensors are limited. The map is mostly hidden behind Fog of War. The target knows you are coming.

Your mission is to navigate the facility, locate SKYE-X, and shadow it persistently within a 70-pixel tracking radius for as long as possible before the mission clock runs out.

SKYE-X will not make this easy.

2. The Enemy — SKYE-X

SKYE-X is not a passive target. It has its own onboard AI with three distinct behavioural layers that activate depending on how close you are:

Behaviour Trigger What SKYE-X Does
Gaussian Random Walk Always active Moves with smooth, unpredictable momentum — never fully still, never fully predictable
Seek Cover You within 80 px Identifies the nearest obstacle and steers toward it to break your line-of-sight. This is its most dangerous capability.
Direct Escape You within 80 px, no cover nearby Flees directly away from you at full speed
Wall Avoidance Near arena edges Soft repulsion keeps it from cornering itself — it will not trap itself for you

The cover-seeking behaviour is what separates good controllers from great ones. SKYE-X actively places rectangular obstacles between itself and you, causing target_visible to drop to False and forcing you back into blind exploration. If your pursuit logic cannot recover from a lost target, your score will stall.

3. The Environment

Arena

Parameter Value
Width × Height 1280 × 900 pixels
Obstacles 22 randomized rectangles (crates, pillars, barriers)
Player Spawn (80, 80)
Target Spawn (WIDTH − 150, HEIGHT − 150) ≈ (1130, 750)
Mission Duration 3000 timesteps (default — configurable via config.py)
Frame Rate 60 FPS

Drone Physics

Parameter Value
Max Player Speed 3.5 px/timestep
Player Collision Radius 10 px
Max Target Speed (Normal) 3.5 px/timestep

Note: Speed is enforced — if your output velocity vector exceeds MAX_PLAYER_SPEED, it is automatically clamped to that magnitude.

4. Sensor Suite — What You Can See

Your only window into the world is the get_sensor_data() return dictionary:

sensors = {
    "player_x":        float,   # Your X coordinate
    "player_y":        float,   # Your Y coordinate
    "lidar_distances": list,    # 36 floats — one per ray (every 10°)
    "target_visible":  bool,    # True if target is within VISIBILITY_RADIUS (150px)
    "target_pos":      tuple    # (tx, ty) if visible, else None
}

LiDAR System

  • 36 rays, equally spaced every 10° (full 360° sweep)
  • Each reading is the distance to the nearest obstacle/wall in that direction
  • Maximum range: 150 px (Visibility Radius)
  • Beyond that range: the map is hidden behind Fog of War

Visibility

  • You can only detect the target if it is within 150 px of your position
  • target_visible becomes True the moment the target enters this radius
  • Once discovered (target_discovered = True), scoring begins

5. Collision Rules

Your drone must avoid:

  1. Arena boundaries — flying outside [0, WIDTH] × [0, HEIGHT] instantly crashes the drone.
  2. Rectangular obstacles — collision is detected using circle-AABB: if your drone's center comes within 10 px (DRONE_RADIUS) of any obstacle edge, it crashes.

A crash immediately ends the mission with your current score — no partial credit.

6. Objectives & Scoring

Score Accumulation

  • +1 point per timestep that the following conditions are both met:
    1. The target has been discovered at least once (target_discovered == True)
    2. Your drone is within 70 px (TRACKING_RADIUS) of the target's current position

Mission Termination

The mission ends when any of these occur:

  • timesteps >= MAX_TIMESTEPS (default 3000, configurable in config.py)
  • Player drone crashes into a wall or obstacle

What Determines a Good Score?

  • Fast exploration: Find the target quickly to start accumulating points early
  • Persistent pursuit: Stay within 70 px despite the target actively fleeing
  • Obstacle awareness: Cut through the maze efficiently without crashing

7. Your Task

Open skye_controller.py and implement the compute_velocity(sensors) function.

def compute_velocity(sensors) -> (float, float):
    """
    Input : sensors dict (see Section 3)
    Output: (vx, vy) — your velocity command for this timestep
            ||(vx, vy)|| will be clamped to MAX_PLAYER_SPEED = 3.5
    """
    ...

A starter implementation with a dual-state template is already provided:

  • State 1 (Exploration): When target_visible == False — explore the map
  • State 2 (Pursuit): When target_visible == True — chase and track the target

You are free to use any approach: hand-tuned heuristics, potential fields, A* path planning, Reinforcement Learning, or anything else.

8. Deliverables

Teams must submit a single GitHub repository containing all of the following:

# File Description
1 skye_controller.py Your implementation of compute_velocity(sensors). This is the only file you modify. Must run without errors on the unmodified skye_env.py and config.py provided.
2 score_proof.png A screenshot of your final simulation run showing the score on screen. Taken at mission end (timestep 3000 or crash).

9. Constraints

Rule Detail
Modify skye_controller.py (your main solution)
May modify config.py — you may adjust parameters like MAX_TIMESTEPS for tuning/testing
Do not modify skye_env.py
Do not Hard-code target coordinates or exploit rendering internals
Allowed Any Python library (numpy, scipy, etc.)

10. Scoring Summary

Outcome Score
Crashed immediately 0
Found target but failed to track Low (< 50)
Found & tracked for most of mission Medium (200–800)
Perfect pursuit + survival High (> 1000)