AI Basketball Training: First, Collect My Misses

COURTSIDE IDEAS · VIGO · OCTOBER 2026

By iJ Atlantic

Everybody wants to see a robot score a spectacular basket. I have a more urgent question: who is going to collect the ball after my shot, whether I score or miss?

I like long-distance shooting, and I like inventing things. These two interests ought to cooperate. A beautiful shot is a pleasure. A ball that rolls towards the other end of the court immediately afterwards has a different opinion about how I should spend my afternoon.

So, when I think about AI basketball training, I imagine a small assistant with a practical role. I would take the shot. The robot would then collect the ball and return it to me, whether I scored or missed. The robot could also announce the score, preferably without sounding too disappointed when I missed.

The basket can stay over there. A robot could bring the ball back.

A robot can shoot. What would help me train?

There is already a serious example of robotic shooting. In December 2024, Toyota reported a 24.55-metre basket by its CUE6 robot. That is an impressive distance. The video below shows the project through Guinness World Records.

I enjoy this kind of engineering. Nevertheless, if a robot came to my training session, I would first ask it to help me take the next shot. It can have its own three-point contest afterwards.

Video: Guinness World Records, featuring Toyota’s basketball robot. Watch the video on YouTube.

My small robots came before the AI labels

In 2019 I presented an instructor robot based on a small wheeled platform, with music and voice commands stored on a microSD card. The app used App Inventor 2. I also presented my Object Finder project, based on Bluetooth signal strength, or RSSI.

Neither project was a finished basketball rebounder. They gave me a concrete starting point: movement, signals, simple instructions and a machine that could participate in an activity. Giving a robot a voice is entertaining. Giving it a useful job is more interesting.

My Object Finder robot, with its LEDs illuminated, from the 2019 project.

The assistant I would actually use

My first version would need to locate a stationary ball, approach it, stop correctly and bring it back. A missed shot can finish anywhere. Even after a successful basket, I still have to collect the ball; it does not return to my hands by itself. Bluetooth signal strength by itself does not tell a robot where an ordinary basketball is. Ball detection, collection and safe navigation would be additional engineering tasks.

I would first test the robot in an empty area of the court, keeping its speed low and using a clearly marked control to stop it whenever necessary. Later, I would program the robot to guide my training with spoken instructions: “Take ten shots from this position. Change sides. Rest. Repeat.” I would perform the exercises, while the robot would give the instructions and collect the ball. The robot would also record my shot attempts and successful baskets, and I would be able to correct its count if it made an error.

As for the robot’s personality, I would like a setting to adjust the sarcasm in its voice. After a terrible shot, I could choose whether its “nice attempt” sounds encouraging or ironic.

Try the ball-collection experiment

Before building anything, we can examine one small part of the idea. Choose a session below and estimate your average collection walk after a miss and after a basket. Include both the walk to the ball and the return to your shooting position. Then decide what percentage of balls a hypothetical assistant would return after each outcome.

Who returns the ball after your shot?

Every shot leaves a ball to retrieve. One very willing robot could help.

540 mWithout a helper

270 mYour remaining walk

270 mCollection walk avoided

For example: 60 attempts at 50% give 30 baskets and 30 misses. At 8 metres to collect each basket and 10 metres to collect each miss, the total collection walk is 540 metres. If a helper returns half the balls after both outcomes, your share becomes 270 metres.

Concept animation: a small robot collects and returns a basketball after a shot.

A missed shot is illustrated. The helper would also return the ball after a basket.

Expected baskets = attempts × accuracy ÷ 100. Expected misses = attempts − baskets. Collection walk = misses × walk after a miss + baskets × walk after a basket. Each distance includes going to collect the ball and returning to the shooting position. The helper percentage applies equally to made and missed shots. Distances and helper performance are assumptions, not measurements. The animation does not model these distances.

This calculation says nothing about how many extra baskets I would score. Walking is exercise too. The attraction is being able to choose the purpose of the session: practise the release, practise movement, or recover the ball. I would rather choose what to practise than keep interrupting my shooting to pursue a ball that seems determined to escape to the other end of the court.

The player still has to play

A camera and an AI system could be another assistant, helping me review a sequence and compare what I intended with what I actually did. I would want the original video beside any analysis. An attractive explanation is not enough if the system has lost the ball behind another player.

I am interested in a modest first experiment: one player, one ball, one collection task and a result that can be checked. If that works reliably, we can add complications. Basketball will provide plenty without our assistance.

Would I eventually like to see these ideas in a much larger basketball project in Vigo? Of course. But even the most ambitious project needs somebody to return the ball after the first shot. Which job would you give the robot: return the ball, organise the drill or analyse the video? Tell me in the comments.

Where this idea comes from

My Object Finder robot using Bluetooth RSSI is the practical starting point for the discussion about signals and movement. It shows the earlier project behind the illuminated robot in the photograph.

My Instructor Robot by π9HMB develops the other half of the idea: a machine that participates in sports practice through music and voice commands.