Kiwi Charge — Autonomous EV Charging Robot

Kiwi Charge addresses the last-mile friction in EV adoption: plugging in. The robot autonomously navigates to a parked vehicle, identifies the charge port using computer vision, and physically connects the appropriate charging connector — NACS, CCS, or wireless — without any driver interaction. A companion mobile app lets a driver request a charge, track the robot's status, and get notified when charging starts and completes.

The mechanical design of the end effector and connector flange was a central engineering challenge, requiring precision tolerancing for reliable mating across real-world charge port variations and orientations. Two connector standards were engineered to share a single arm-mounted flange, reducing tooling changeovers in the field.

Custom end effectors were designed in SolidWorks for both NACS and CCS connector standards, sharing a common flange and cable-routing spine so either connector head can be mounted on the same arm. Each end effector integrates a force-compliant mounting that accommodates small misalignments during insertion — critical for real-world reliability where vehicles are never parked perfectly.

Wireless charging hardware for the pad-based system is also in development; specifics are being kept under wraps while a patent application is pending. Connector retention and release mechanisms were designed for repeatable actuation by the robot arm, and all components were prototyped via 3D printing and iterated against physical connector samples before transitioning toward machined, production-intent parts.

The insertion pipeline fuses a stereo depth camera mounted at the wrist with a force/torque sensor in the flange. Port pose is estimated from the depth stream, and the arm is driven toward it using a stack of inverse-kinematics solvers chosen per approach phase — coarse IK for the initial reach, then a finer, damped-least-squares solve for the final approach where accuracy matters most.

Once the connector makes contact, control hands off to a force-feedback loop: PID and admittance control keep contact forces within safe bounds while the arm searches out small pose corrections, letting the system tolerate real-world misalignment instead of relying on a perfect approach trajectory. The mobile app talks to the robot's onboard controller to queue a charging session, report robot/vehicle state, and confirm session start and completion to the driver.

Successfully demonstrated autonomous connector mating with both NACS and CCS charge ports across multiple real vehicles, including Tesla Model 3 and Model Y, with high port-detection accuracy and zero unintended contact events during testing. The end effector design achieved reliable insertion across a range of port heights and lateral/angular misalignment typical of real-world parking.

The project — and the perception, insertion-control, and charging-stack work behind it — was presented at a Clean Energy Zone showcase at Toronto Metropolitan University's Centre for Urban Energy to industry professionals and researchers.