Let me paint you a picture of what I want the future of transportation to look like:
I’m at school and want to get to the soccer field two miles away. I open up an uber-like app and hail an autonomous bicycle, which promptly arrives at my location in only a few minutes. Once I get on, I am fully in control. Autonomous collaborative balancing is not something I need in my life. But as soon as I arrive at the field, I hop off and the bike goes off to the next rider. No clutter on the sidewalk, time and money well-spent.
Why a bicycle? Well for one thing their lightweight frame and only two contact points with the ground make bicycles the most efficient ground vehicle that can successfully navigate a city. They are able to move as quickly as a car in an urban setting, but their lower profile makes them much more nimble and less dangerous in the case of a collision. Humans are also able to ride them safely and intuitively, which opens new opportunities unavailable to other small robot delivery vehicles. In short, autonomous bicycles fit in the perfect niche of efficiency, safety, speed, and human-friendliness that make them highly desirable in a community.
Use Cases:
- Very cool personal bicycle
- Convenient, clutter-free bike-share programs (with all the advantages of autonomous vehicles)
- Faster, more efficient robot delivery services (better than Starship or Kiwibots)
- City-scale inspections/mapping/security
- For academics in robotics, the fact that this vehicle is non-trivial (harder than a cartpole) but still analytically approachable (not as bad as a humanoid) make it an excellent platform to test control algorithms both in hardware and in simulation.
Now you may be thinking, I like this idea, but wouldn’t the bicycle just fall over and get stuck? That is certainly the primary limitation of a normal bicycle, but not so with this design! By putting a joint roughly coaxial with the where the bottom-tube would normally be, this bike is able to passively rest in a half-upright position (you kind of have to see it to believe it). With a good controller, it can swivel into the upright position, and once it is moving balance is easily obtained through steering, as a human would do. Notably using the joints this way is entirely novel as far as I’m aware. Existing autonomous bikes rely on flywheels (heavy and clunky), and/or do not include a mechanism to stand up from a fallen down position.
With all this in mind, my goals for this project are to design a bike that can (1) automatically right itself from any position on the ground, and (2) drive autonomously (without person onboard), so that riders can summon it and it can do deliveries. The primary challenges are to design hardware that is physically capable of righting itself, creating a balancing controller, and finally building a perception, planning, and control stack that can effectively navigate autonomously.
As of August 8, 2024 I’ve completed a physics (pybullet) simulation and a small physical model that prove the concept, but have yet to build the larger system. Below is a report with more of the math, still a work in progress. If you’re interested in making one of these yourself, please cite me and reach out! I’m happy to help.
Robocycle-Report