Map data © OpenStreetMap contributors · terrain: AWS Terrain Tiles (SRTM)
Alen Hsiao
I study how intelligent systems learn,
and build games and tools people can use.
探 tàn — to explore
Explore Taipei
Where I live and study, modelled small: the basin at dusk, 101, the river, NTU.
Scroll to fly across · drag to look around · choose a place to fly there

01 · 作 zuò — to make
Work
Robot learning at NTU, and the AI systems I build around people.
IllustrationOffline goal-conditioned RL Learning to reach a goal from a fixed dataset, in NTU’s Embodied AI Lab.
IllustrationExhiBeat The AI agent of a Taipei exhibitions-and-events platform: it finds shows, plans the day and suggests a stay nearby.
Pick a goal · interactive 3D sketch
Goal-conditioned control in miniature: name a goal and the arm works out how to reach it. Or point and click to drive it yourself.
Focus this, then arrow keys move the gripper and Enter picks up or puts down a block.
The 3D sketch loads as you come near.
Modelled in three.js for this site · not lab footage, not a learned policy
All work
02 · 遊 yóu — to play
Play
Games made to be played together, in a browser.
Camelot Nights Social deduction for 5–10 players, with bots, voice chat and cinematic reveals.
Bullheads The 104-card “don’t take the sixth card” game, for 2–10 players in a browser.
Camelot Nights · the hidden roles
Tap a card to reveal a role · Illustration





