For the 2025- 2026 school year, I am building a jumping rover for NASA HUNCH. The main objectives for the rover are to repeatably jump >3ft, stream video, and be remotely controlled. The rover will be controlled by either an ESP32 or an NRF52840 microcontroller. The rover will be powered by a 3 or 4s LiPo battery. I have already drawn the linkage by which the rover will jump, requiring only one driven link for each of the two legs.

Overview

Ambitously, the rover will be bipedal, with 2 indivually controllable legs. I plan for each of the legs to have 2 degrees of freedom: the main brushless motor that moves the linkage, and maybe a servo at the knee that acts as a calf to actuate the heel. The main motors will each be controlled by a 15 amp speed controller made for rc boats, as the typical ones for drones are not bidirectional, which is required for my jumping mechanism. There will be a singular 750mah 3S (11.1V) LiPo battery. The microcontroller will be an ESP32; however, this requires 3.3V, so I will make a voltage regulator PCB to step down the voltage from the battery to 3.3V for the MCU and other peripherals, e.g. camera, encoders. All strucutral components will be either 2mm or 1mm sheet metal due to tolerances & strength to weight. 5075 alloy was chosen due to its elasticity (the more common 6061 is more brittle and prone to cracking, not ideal for sudden impact of landing). Arterial, not highly load bearing components will be 3D printed pla. The main motors are 1750KV drone motors, which need a ~700:1 gear ratio for adequate output rotation speed and torque. The gears are made of laser cut delrin, as it is low friction and strong. The rover will have a camera module, like a seperate 5.8 ghz camera module from FPV drones, whose video stream can easily be accessed via a usb adapter which practically acts as a normal webcam to the computer, but instead receives video over 5.8 ghz. Right now its looking like it'll weight ~350 grams.

Energy Storage

Each leg has 2 springs, one compression spring and one torsion spring. My greatest intent for either spring is that they dont impede on the extension of the leg when stretched passed their equilibrium point. The compression spring works pretty straightforwardly: when the leg is compressed, the spring compresses, storing energy. When the leg is extended, the spring extends, releasing energy. The torsion spring is a bit more comples. The input of the linkage is a gear, driven by the gear box. This input gear is comprised of three layers, 1 layer is a link of the leg linkages, and the other 2 are used to kind of sandwich the torsion spring in parallel between the gearbox and input link. When the spring is compressed past the neutral point, one end of the torsion spring pushes against the gold standoff, compressing the spring and storing energy prior to a jump. When the leg is extended the torsion spring presses against nothing, thus not impeding the extension of the leg. The torsion spring is primarily used to store energy for the initial part of the jump, while the compression spring is used to store energy for the latter part of the jump. This is because the torsion spring can be compressed more easily at small angles, while the compression spring is more effective at larger displacements.

Further Ambitions

I thought it would be cool if the computer controlling the rover could parse the video input stream into some yolov8 detection model to detect obstacles, people, landmarks, etc. This is actually pretty feasible, as yolov8 is easy to train and performs well. Note: this is not running on the esp32, but the computer that the aformentioned 5.8ghz usb video adapter is plugged into. Further, I think it would be cool for some natural language component. Imagine I type into a prompt in the computer that I want the rover to move 200 feet forwards or go look at the big rock. The computer could parse this command, and use some pathfinding algorithm to determine how to get there, sending low level commands to the esp32 to move the rover accordingly. This is a bit more far fetched, but I think it would be a cool challenge to implement.