Show HN: Lessons from building a SIMA2 style agent in Roblox
Frisson Labs is exploring the development of a SIMA2-style agent within Roblox, aiming to create engaging AI companions. The project focuses on enabling the agent to learn from human gameplay and improve its actions through experience. Initial tests with the game Slime RNG revealed both successes and challenges in achieving human-like movement and decision-making.
- ▪Frisson Labs is building a SIMA2-style agent in Roblox to create fun AI companions.
- ▪The agent learns from human play and uses screen observations to make decisions.
- ▪Initial tests with the game Slime RNG showed the agent performing player-like actions but also struggling with navigation.
Opening excerpt (first ~120 words) tap to expand
What if we made SIMA2 from Temu By Rohit Swamy | May 20, 2026 | Roblox agent field report At Frisson Labs, we want AI companions that are actually fun to play with. Playing the game is the first step. Before an agent can feel present, useful, or worth keeping around, it has to handle the basics of being in the world. That means it has to: read the screenfigure out what to dorecover when it gets stuck SIMA2 is the best public example for that direction: agents that learn from human play, act from screen observations, use keyboard/mouse controls, and use reasoning for longer-term goals. Google DeepMind's SIMA 2 overview video. We are a small lab, so we are not going to outspend Google on game-agent training.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Frisson-labs.
Discussion
0 commentsMore from Frisson-labs
