Strengths:
- Hotspot detection and pose estimation at low speeds: Salus is good at detecting door handles and light switches when moving at low speeds.
- Autonomously navigating a sparse classroom environment: Salus is capable of navigating a sparse classroom environment while avoiding obstacles reliably.
- Spray coverage: When Salus is spraying hotspots, the spray has plenty of coverage to disinfect the entirety of the hotspot.
- Cost: The raw materials for Salus cost less than $10,000, which is much less than competing robots in this market.
- Traceability of disinfected hotspots: Salus reliably logs each hotspot that is disinfected while operating.
Weaknesses:
- Robustness of the mobile base: Throughout the two semesters working with the Marvelmind robot we experienced a number of hardware/electronics failures including multiple shorted PCBs and blown motors.
- Mobility of the base: The Marvelmind robot has two driven wheels and four caster wheels. The caster wheels often times get stuck, and the two driven wheels do not have enough torque to overcome them. The base is also large and heavy, making it difficult to overcome the inertia of the robot.
- Robustness of the manipulator-sprayer mechanism: The manipulator-sprayer mechanism is heavy, and it undergoes a lot of shaking/vibrations while Salus moves. This shaking caused the shaft couple between the motor and the ball screw to break during FVD Encore.
- Hotspot detection at a distance: Salus was good at detecting hotspots when they were within a couple meters, but struggled at larger distances. This is partially due to the angle at which the camera is mounted, which causes door handles and light switches to not be in the camera’s field of view when more than a couple meters away, and partially due to a lack of training images at larger distances.
- Hotspot detection and pose estimation while moving: Salus struggles to detect hotspots while Salus is moving due to camera blur, and the pose estimates are less accurate because Salus’ pose data is not synchronized with the camera data, leading to errors in the transforms of the hotspot pose estimates from the base frame to the world frame.
- Obstacle avoidance in a cluttered environment: Due to Salus’ large chassis, the sprayer sticking out well past the chassis, and mobility issues with the base, Salus struggles navigating in cluttered environments.
- Ability to autonomously handle faults: Watchdog was designed to warn the user of any errors occurring in the system and block actuation if necessary, but it had little ability to intervene with any faults. A greater emphasis on developing advanced capabilities in Watchdog could improve Salus’ ability to handle faults autonomously.