Overview

HandAid is a 4-degree-of-freedom robot arm built to help hospital patients with limited mobility feed themselves. For someone recovering from surgery, living with a spinal cord injury, or dealing with tremors, eating usually means waiting for a nurse or family member. HandAid aims to give some of that independence back with a small bedside arm that brings food from the plate to the patient's mouth.

The arm is a serial chain of four revolute joints, driven by daisy-chained Dynamixel servos mounted on 3D-printed links. I drafted the initial mechanical design in SOLIDWORKS, wrote a Julia package that solves the arm's forward and inverse kinematics from its Denavit–Hartenberg (DH) parameters, built a MATLAB app to simulate and visualize its paths, and wrote the servo position control with the Dynamixel SDK, including PS4 controller input for teleoperation.

Background

Feeding assistance is one of the most common and most time-consuming tasks in inpatient care, and it's one where losing independence hits hard. Commercial feeding robots exist, but they're expensive, bulky, and often limited to fixed scoop-and-lift motions that don't adapt to different plates, foods, or patient positions.

Being next to a hospital bed puts real constraints on the design:

• Safety: the arm works near a patient's face, so its motion has to be slow, predictable, and low-force.
• Footprint: it has to fit on an overbed table without getting in the way of care.
• Cost and serviceability: printed parts and off-the-shelf servos keep it cheap and easy to repair.
• Simple control: the patient, or a caregiver, needs to be able to operate it with minimal training.

Design

The arm uses four revolute joints. A base joint swings the arm between the plate and the patient. Shoulder and elbow joints set reach and height. A wrist joint at the end of the forearm turns the 3D-printed utensil holder to tilt the spoon.

The links are 3D-printed, which made it fast to iterate on lengths and joint housings. Each joint is driven by a Dynamixel servo, mostly MX-28AR units. The servos are daisy-chained on a single bus, so the whole arm runs on one cable, and each servo provides closed-loop position control and reports its own position.

Kinematics

The arm is described with standard DH parameters, one row per joint. From those, the Julia package builds each joint's homogeneous transform and multiplies them along the chain:

• Forward kinematics: given the four joint angles, compute the pose of the spoon.
• Inverse kinematics: given a target spoon position (the plate, or the patient's mouth), solve for the joint angles that reach it.

# TODO: paste the DH table and forward/inverse kinematics functions from the Julia package

Before running anything on hardware, every path was checked in a MATLAB app that animates the arm along a planned trajectory and plots its moving coordinate frames. That catches unreachable targets, joint-limit violations, and awkward elbow flips on screen instead of on the robot.

Validated joint trajectories are sent to the servos as position commands through the Dynamixel SDK. A PS4 controller mapping lets an operator drive the arm directly.

Results

The prototype pairs the printed arm with the Julia kinematics and the MATLAB simulator: trajectories are planned and checked in simulation, then sent to the servos as position commands.

Conclusion

HandAid brought my mechanical, math, and software work together in one project. The kinematics are textbook. Most of the effort went into the details around them: keeping simulation and hardware consistent, picking DH frames that keep the math manageable, and designing printed parts that hold up under servo loads.

Future directions:

• Closed-loop targeting: add a camera so the arm finds the plate and the patient's mouth itself, instead of using preset positions.
• Compliance and safety: use servo current sensing to detect contact and stop, important for any device near a person's face.
• Accessible input: add voice, sip-and-puff, or head-tracking controls for patients who can't use a gamepad.