Home▸Robotics & Kinematics▸Object Capture 2D — Sensor-Guided Dwell Grasp

Object Capture 2D — Sensor-Guided Dwell Grasp

Interactive 2D companion to the 3D Object Capture RL environment: a single differential-drive rover with five raycast sensors learns to steer around obstacles, slow down over the target, and dwell inside the capture zone long enough to complete a genuine object capture.

Robotics & Kinematics2DModerate60 FPS📱 Mobile-adapted⇄ 3D version

The 3D model trains a swarm of massless points to glide toward goal orbs through a potential field. This 2D companion instead drives one differential-drive rover with five raycast sensors: it has to steer around obstacles it can only sense as ray distances, brake as it nears the target, and hold still inside the capture ring for a set dwell time before the object actually counts as captured. Watch the reward curve tighten as the exploration noise (ε) decays episode over episode — the same training-convergence signature as the 3D dashboard, produced by a completely different rover-and-sensor mechanic.

⚙ Under the hood

Interactive 2D companion to the 3D Object Capture RL environment: a single differential-drive rover with five raycast sensors learns to steer around obstacles, brake over the target, and dwell inside the capture zone long enough to complete a genuine capture — a graze-and-leave earns nothing. Exploration noise decays over training time, tightening the live reward curve.

Canvas 2DReinforcement LearningRaycast SensorsDifferential DriveObstacle AvoidanceReward CurveRobotics

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install