Yess.. I have finally completed the whole design. This time everything was well thought out. Instead of normal wheels I have decided to use mecanum wheels for omnidirectional movement.

Also there will be two cameras instead of one. The first camera is on the chassis as eyes. The eyes give depth perception so the robot can judge how far away an object is without a depth sensor. The second camera will be on the arm head for object detection.

Electronics and the battery sit low at the back of the chassis as a counterweight to balance the arm mounted at the front, keeping the center of gravity stable even when the arm is fully extended.

The arm has 6 degrees of freedom plus a gripper claw. Serial bus servos give position, voltage, and temperature feedback at every joint, which means the arm always knows exactly where it is, no guessing, no drifting. Inverse kinematics handles all the math of translating “go to point (x, y, z) in space” into the correct angles for all 6 joints simultaneously, so I can give the robot a target position in plain language and it figures out how to get there.

Voice commands flow through a full natural language pipeline: a wake word detector listens constantly for “Hey Dum-E,” then Whisper transcribes what I say, then an LLM interprets the intent and converts it into structured robot commands, things like pick up the red block, move left 10 centimeters, go to the charging station, lift your arm, or grab that.

The full pipeline from voice command to physical grab is: wake word, then speech to text, then LLM intent parsing, then object detection, then coordinate mapping, then inverse kinematics, then servo motion. All of this runs on the Raspberry Pi 5 with an Arduino handling the real-time motor and servo PWM timing underneath, because the Pi’s Linux scheduler isn’t real-time and would cause jittery motion if it controlled servos directly.