Inspiration

I seek an improved quality of life, and I've noticed that a person's quality of life strongly influences their productivity. Those with better quality of life experience improved mental health and tend to outperform those with poorer quality of life. Research has also shown that people in natural settings are more productive. To maximize nature's presence, my motivation is to clean up the streets and introduce more green spaces. Here are some articles for reference: 1, 2, 3.

Furthermore, like many others, I desire a long and healthy life. Issues such as construction in downtown Berkeley, litter on the ground, and waste entering the ocean harm us. While policy changes can help minimize these problems, another controllable approach is to automate the cleanup process. I envision robots patrolling the streets and waters, cleaning and protecting our environment, thus enabling us to lead healthier and happier lives.

What it does

Ideally, these robots would be designed to appear as cute and friendly as possible to provide comfort rather than causing alarm. In "idle" mode, they would patrol a designated area, limited to that specific region, and clean up any litter they encounter. The cleaning process can be simplified into three steps:

1) Using senses like touch (e.g., detecting the texture of gum or waste on concrete), smell (e.g., sulfur or carbon detectors), and sight (e.g., edge detection, infrared), the bot would differentiate between objects that belong and objects that are waste/hazardous.

2) The bot would safely remove the identified waste, employing appropriate methods depending on the type of waste encountered (e.g., scrubbing and sanitation for waste, neutralizing battery acid with a base liquid, safely handling sharp objects by breaking them down and disposing of the components responsibly or repurposing them).

3) The bot would continue its patrol.

In addition, the project includes a human interaction aspect, which serves as a primary source of revenue. These bots would be capable of conversing with people when ordered for a specific service. For example, a local restaurant could request the bot to clean the bathroom. The bot should be able to listen to, obey, and respond to verbal directions. Censorship and restrictions would be implemented in the bot's AI to ensure safety.

When necessary, the bot would recharge at a designated station and be replaced by another bot until fully recharged. Ideally, the bots would utilize an existing power source, such as an electric vehicle charger, in a secure environment.

How it will be built

The prototype will be constructed using a Raspberry Pi device, but the final product will incorporate stronger components. The initial bot will consist of a Raspberry Pi device connected to a robotic arm, a camera, a microphone, traversal motors, wheels, and a waste bin. It will be provided with a general map of the area to follow a predetermined patrol path. The camera's vision will help the bot avoid harm and detect objects to pick up.

When a service is ordered, the user can submit directions through a downloadable program or via voice commands. These commands will trigger an onboard GPT-3.5 model (specifically, a completion model rather than a conversational one) to generate steps for completing the task (ideally, it would trigger a GPT-4 model response , but Raspberry Pi can't handle that, so the trigger would be a call to a remote server or simply replace the Raspberry Pi with better hardware). The generated steps will include recurring keywords that are pre-programmed to map to specific actions (e.g., sweep, wipe, scrub, rinse). The bot will parse the generated steps and perform the corresponding actions at each keyword. The job will be considered complete once the user is satisfied. However, the GPT-3.5 model may forget some steps if the task takes too long, in which case Pinecone could be utilized. Additionally, Hume can assist in conversation to determine user satisfaction.

Challenges we faced

We encountered challenges with funding and the learning process. Initially, I need to secure a well-paying job to save up and build a legitimate robot. Additionally, I have been learning new technologies required for implementation.

Accomplishments we're proud of

Through this project, I have found direction, renewed motivation, and gained a better understanding of where I am in life right now.

What we learned

We have discovered that we can build impressive machines using Raspberry Pis and Arduinos.

What's next for Project Ash

The next steps for Project Ash involve securing safe funding, collaborating with material science experts, and integrating the project into the public sphere.

Built With

Share this project:

Updates

Submission history