Avionics for sub-gram aerial robots

The aim here is to achieve sensor autonomy to obtain state feedback necessary for closed-loop attitude, altitude and lateral velocity control. A sensor suite comprising a gyroscope, optical flow sensor and a laser rangefinder (or a barometer) provides a measurement model that satisfies the observability criteria. Kalman filter variants are suitable for sensor fusion, and the estimates can then be fed into a controller such as PID or LQR.

While inertial measurement units (IMU), laser rangefinders and barometers are available at a low enough weight (~ 15mg), commercially available optical flow sensors widely used in conventional drones are too heavy (> 120 mg) for this work. To solve this, we utilize a monochrome global shutter imaging sensor paired with a sufficiently small lens and implement a highly optimized optical flow algorithm based on the Lucas-Kanade method.

Key specs:

• weight including microcontroller: 97 mg
• dimensions after folding: < 1 x 1 x 0.5 cm^3^
• estimation (driven by IMU sampling frequency) at 500 Hz
• customized optical flow estimation at 100 Hz
• Low-latency real-time wireless data transmission

Prior published work: (Yu et al., 2025), (Talwekar et al., 2022)

Pushing further: gyroscope free visual-inertial flight control for 10-mg robots

Due to the physics at the small scale, under free-flight, accelerometers can be used to estimate the drag force acting on the robot, which in turn can provide lateral velocity information. Combined with a camera, we can formulate an estimator for attitude and velocity control inspired by the wind-vision sensor fusion in the fruit fly Drosophila melanogaster. Work published in Science Robotics (Fuller et al., 2022)

Autonomous mobile robotic jigsaw

Bladed tools such as jigsaws are common tools for wood workers on job-sites and in workshops, but do not currently have sufficient autonomous hardware or path planning algorithms to enable automation. This work developed a system of an autonomous robot and a path planning algorithm for automating jigsaw operations. The path planning algorithm converts complex shapes into paths for the jigsaw, drill holes, and traversal movements for the robot.

The robot can autonomously:

• drill holes
• align itself with the drileld hole to safely insert the jigsaw
• Cut the part by following the defined trajectory

The robot localizes itself using feedback from an HTC Vive Lighthouse system, and trajectory tracking is done online using an MPC framework. Communication between ground PC and on-board computer happens over ZeroMQ.

Published work: (Zhao et al., 2021)

Sensing wearables for biology

The natural world serves as an inspiration for many applications in technology. These incredible adaptations have led to the development of many approaches in all domains of sensing, control, actuation, design and more in robotics. In turn, these advancements prove to be useful in understanding biology itself. While imaging techniques are incredibly useful, there are certain phenonmenon that can only be studied by deploying sensors on the organism itself. The biggest priority in deploying any payload on these organisms is their comfort; we do not intend to stress them or restrict their motion in any way. Further, small organisms such as insects and birds have challenges due to their small size and how much they can carry. These are very similar to the SSWaP constraints we face in developing insect-scale robots, therefore the solutions we come up here are well-suited for biological applications as well.

Efforts in developing miniature sensing systems has the potential for opening up a new exploration space for biologists by providing them new tools to add their vast array of amazing research techniques. For instance, my work on lightweight sensing systems is being used to study the flight dynamics of hummingbirds and moths. Key specs of these backpacks:

• features a 6-axis IMU, microcontroller with 2.4 GHz radio, hall-effect sensor as a switch
• weight under 300 mg including battery
• dimensions around 6 x 7 x 10 mm^3
• two way communication over bluetooth low-energy (BLE) for configuration and data streaming
• IMU data is polled and streamed in real-time at 500 Hz, can be increased as needed
• run-time of over 10 mins at 500 Hz with a 0.8 mAh battery

All animal handling done by qualified personnel with required permits.

I am also actively working on building a compact vision system that can be attached to these organisms to help researchers gain insight into their behavior patterns based on their perception of the environment.