Revisiting Wirepod for Vector robots

We tried to re-install wirepod and check out the integration with OpenAI

We first wrote about Wirepod soon after it was developed last year. We were impressed by the engineering and the effort that had been put behind it. Wirepod truly liberated owners of the Anki Vector robot from the clutches of DDL. Vector got a new life… and endless possibilities opened up to use Vector in various domains.

Wirepod today

One year later, Wirepod has become critical for whichever Vector robots still remain in use and in the market. Meanwhile, DDL seems to flop as a company, DDL servers have been down for almost a month, which means no-one can really use Vector (other than programming it via the Vector SDK). You cannot even set up a new Vector via the Vector App. The only recourse is to use Wirepod.

In the meanwhile, Wirepod use has grown a lot in the Vector community. A look at the Vector subreddit shows a lot of questions and issues people have faced regarding the installation of Wirepod. So we decided to retry Wirepod from Scratch for our two Vector robots: an OSKR robot and a production robot. Here is our take on how to solve the different issues that came up.

1. Have Vector and Wirepod on the same network (subnet). Wirepod will not work if you install it on a cloud virtual machine as an example. The best way is to check Vector’s IP address (place Vector on his charger, and move the lift arm up and down) and check the IP address on the machine Wirepod is installed on. They should be on the same subnet (As an example, my Vector has an IP address of 192.168.31.18 and the Ubuntu server wirepod runs on has the IP 192.168.31.7, so they are on the same subnet). Almost all successful installations I know of installed Wirepod on Ubuntu. There is a guide to installing wirepod on WIndows 10/11, but I have never tried it.

2. Use the Chrome browser to update Vector’s firmware. The browser needs to pair with Vector using a bluetooth connection. Only the Chrome browser is supported.

3. Leverage Vector’s discord. Vector has a discord called “Vector and Friends”. I am not posting a link here, because the invitation link expired. Once you join discord, you should be able to find it pretty easily. The discord has a channel called #wirepod-help where you can find many questions answered. The lead developer of wirepod hangs around the forum (with the username Wire) and is very active there.

4. Consider buying an off-the-shelf unit. There is a plug and play server available for wirepod. From what I have seen, the seller is very talented and helpful. The off-the shelf unit minimizes the geeky things you have to do to install your wirepod server (Especially good if you completely new to computers and operating systems). Disclaimer: I haven’t tried this unit myself.

New features in wire-pod

Since I first used it, wirepod has grown a lot in many ways.

1. Easy web interface: I remember that my first attempt with wirepod was clunky, I had to copy over certificates using commands, and at that time it only worked for OSKR robots. Now wire-pod has a friendly web interface to do configurations. Although there are still a few technical steps, they are much easier now. And it works for both production and OSKR robots.

2. Open AI Integration: One of my favorite parts of Wirepod is its integration with OpenAI for the knowledge graph. This is great because now you can utilize ChatGPT to answer questions that you ask Vector. All you need is an account from Open AI and the API key. Open AI gives one a few dollars of credit to start utilizing their service. With the help of a free OpenAI account, I could create the following demo video on how Wirepod + OpenAI integration works.

Conclusions

Wirepod offers a basic framework to free your Vector robot from the clutches of DDL and its servers. But it doesn’t stop just there, there is ample room for you to experiment and do your own stuff. You can build custom intents: an easy use case is having Vector respond with your favorite animation for a keyphrase that you choose. You can also experiment with the knowledge graph, such as using some other service that offers Large Language Models (LLMs) to answer questions rather than using OpenAI (Note: this one would need some code changes). We just hope that more people try and experiment with Wirepod… Vector is just so much better using it.

If you have had experience with Wirepod, and have some tips.. please do leave a comment at the end of the post.