A ragdoll physics pong clone made in two days in 2017 updated in 2018 to feature machine learning and network play.
Thanks to it's simplicity, SOCC happens to be a useful device for me to use to expand my skill-set.
These two updates: 0.1.2, which added imitation-trained opponent A.I. through the use of the ML-Agents plugin for Unity, and 0.3.3, which brings online multiplayer as well as a bevy of other improvements. These improvements include customizable controller support, as well as other QoL improvements.
There are 3 difficulty settings, with each bot trained slightly differently. The Easy bot was trained against a human opponent for 17,000 steps, the Medium bot for 30,000 steps, and the Hard bot was trained overnight against the Medium bot.
Online play utilizes the Unity matchmaking service to allow up to four players to join a lobby and square-off in 1v1 matches through a variety of different rotation modes.
For the full patch notes, see the attached link.