AI
Autonomous agents
Design an artificial agent capable of perception, decision-making and real-time behaviour inside a networked Unity FPS environment.
ResearchUnity FPS · Human-Like AI Challenge
Can an AI
fool a human?
Build an autonomous bot for a Unity first-person shooter that behaves so naturally that human players cannot tell it apart from another person.
UNITYModern FPS environment
FPA + TPAHuman judgement
HHuman-likeness
01 / The challenge
Not the best bot.
The most human bot.
BotPrize changes the target of game-AI evaluation: the goal is not simply to win, but to behave in a way that is indistinguishable from a human player.
?
Turing Test in a game
Human judges play alongside bots and decide which opponents are people and which are controlled by an artificial agent.
Human-centred AIU
Open Unity platform
The modern Unity-based environment lowers the entry barrier and supports learning-based, hybrid and cognitively inspired approaches.
Accessible02 / The game
A modern arena for the
classic BotPrize idea.
The 2026 edition reimplements the concept using the Unity FPS Microgame, modified for networked multiplayer and AI-agent integration.
The original question
BotPrize revisits the question associated with Alan Turing: can an artificial agent behave in a way that is indistinguishable from a human player in a fast-paced FPS?
“Can an artificial agent behave in a way that is indistinguishable from a human player in an FPS game?”
Unity FPS Microgame
Built for experimentation.
Unlike the original closed, legacy environment, Unity provides a modern, extensible development environment and room for a broad range of AI techniques.
03 / Evaluation
Two perspectives.
One question.
The submitted agents are evaluated through two complementary phases: direct first-person judgement and anonymized third-person video judgement.
On a game server, a human player is paired with each submitted bot. Judges play and vote on whether observed players are human or artificial.
Participants record and share anonymized gameplay videos featuring bots and humans. Judges then vote on which players are human.
Human-likeness
H = (FPA × FPWF) + (TPA × TPWF)
For the 2026 edition, the published weighting factors are FPWF = 0.5 and TPWF = 0.5, giving equal weight to the two assessment perspectives.
04 / Rules & dates
The road to
CoG 2026.
The competition modernizes the original BotPrize concept first presented at IEEE CIG 2014 and connects it with current research on human-likeness and beyond-score evaluation.
15 APR 2026
Registration closes
Final date published on the competition site for registration.
15 JUL 2026
Approach submission
Deadline for submitting the proposed AI approach.
END JUL 2026
Evaluation phase
Competition evaluation of submitted individuals/agents.
CoG 2026
Results
Results are announced during the CoG 2026 competition session; finalists are announced in advance.
05 / Why BotPrize?
A benchmark for
human-like AI.
The competition is positioned as both a research benchmark and an educational environment for Game AI and AI in Games.
01
Beyond win rate
Encourages alternative evaluation paradigms where the quality of an agent includes how human its behaviour appears.
02
Reproducible experiments
A reusable Unity-based environment can support controlled comparisons and repeated experimentation.
03
Education
Accessible enough for university courses and project-based learning around Game AI and AI in Games.
06 / Results
Competition
results.
Humanness score (H) achieved by each participant, calculated as H = (FPA × FPWF) + (TPA × TPWF).
Humanness score (H) by participant
Agents (bots)
Human players
Podium
01
Player 4H = 88% · Universitat Politècnica de València
02
Player1H = 72% · Universidad de Málaga
03
PepeBotH = 42% · Universidad de Málaga
| Rank | Participant | Team / Affiliation | FPA | TPA | H (Humanness) | Type | Verdict |
|---|---|---|---|---|---|---|---|
| 01 | Player 4 | Universitat Politècnica de València | 100% | 77% | 88% | Human | Passed |
| 02 | Player1 | Universidad de Málaga | 100% | 44% | 72% | Human | Passed |
| 03 | PepeBot | Universidad de Málaga | 50% | 35% | 42% | Bot | Not passed |
| 04 | ADANN | Universitat Politècnica de València | 7% | 21% | 14% | Bot | Not passed |
Ready to build?
Enter the arena.
Get the Unity FPS project and start experimenting with your approach.
07 / Organizers
Built at
Universitat Politècnica de València.
BotPrize 2026 is organized by researchers at the Universitat Politècnica de València.
Juan Peralta DonateUniversitat Politècnica de Valènciajperdon@dsic.upv.es
M.-Carmen Juan LizandraUniversitat Politècnica de Valènciamcarmen@dsic.upv.es
08 / Archive
Past
competitions.
Browse previous BotPrize editions, their specific rules and the historical competition results.
Explore the historical archive.
See how bots and their humanness scores have evolved edition after edition, from the original BotPrize to today.