Games for Science by Jessica Hamrick

Transcript

1. Games for Science! Interactive psychology experiments in Python with Panda3D Jessica Hamrick (@jhamrick) jhamrick@berkeley.edu Peter Battaglia pbatt@mit.edu http://www.jesshamrick.com/ http://web.mit.edu/~pbatt/www/ Follow along: jhamrick.github.io/pycon-2014-talk
2. The rise of games -- , May 2012 (emphasis added) Angry Birds games had so far been played a total of 200,000 years, with 300 million minutes of playing time daily. TNW Blog 300 million minutes = 5 million hours = 208,000 days = 570 years ... per day! What if we could leverage the popularity of games to do science?
3. Fun games for interesting science! 1. The question 2. The hypothesis 3. The experiments
4. How do people reason about the world?
5. How do people reason about the world?
6. How do people reason about the world?
7. How do people reason about the world?
8. How do people reason about the world?
9. How do people reason about the world?
10. Outline 1. The question: How do people reason about the physical world? 2. The hypothesis 3. The experiments
11. Brain as a computer Computational cognitive science software (i.e., mind) Computational neuroscience hardware (i.e., brain)
12. Computer models of behavior
13. Physics games
14. “The physics engine in your head” Approximate simulations are not exact Probabilistic simulations have some randomness
15. Why approximate and probabilistic?
16. Why a physics engine? Effective Simulations are sufficiently realistic Efficient Created to run in realtime General-purpose Applies to many situations
17. Outline 1. The question: How do people reason about the physical world? 2. The hypothesis: "The physics engine in your head" 3. The experiments
18. Ok, so how do we test this? We use video game tools as our theories, and in our experiments. Create a computational model using video game physics (e.g., ) Generate stimuli using a 3D modeling program (e.g., ) Run experiments using a video game framework (e.g., ) (All open source, and all with Python bindings!) Perform data analysis using other Python tools: NumPy, SciPy, pandas, IPython, matplotlib, etc. Bullet Blender Panda3D
19. More information about the research I am just going to focus on the experiments, but if you're interested in the research, check out these links: Project page: PNAS journal article: Scientific American article: Fast Company article: http://web.mit.edu/~pbatt/www/PNAS2013/user.html http://www.pnas.org/content/110/45/18327.abstract http://www.scientificamerican.com/article/the-video-game-engine-in-your- head http://www.fastcodesign.com/3022236/evidence/how-our-brains-play- angry-birds-with-the-world
20. Demo: “Which direction?”
21. Demo: “Which direction?” 1:00 https://github.com/jhamrick/pycon-2014-talk/blob/master/demos/which_direction.py
22. How does it work? 1. Create graphics models to depict the objects 2. Display the models with a rendering engine 3. Add physical dynamics with a physics engine 4. Link graphics and physics together We'll go through a conceptual overview of these steps. If you want to learn how to actually code it, you can find the code here: Panda3D also has a great "Hello, world!" tutorial here: https://github.com/jhamrick/pycon-2014-talk/blob/master/demos/tutorial.py https://www.panda3d.org/manual/index.php/A_Panda3D_Hello_World_Tutorial
23. Step 1: Create a graphics model 3D objects are typically created in a modeling program like (which is Python scriptable!) Blender
24. Step 2: Set up graphics Panda3D is a C++ game library with Python bindings
25. Step 2a: Set up the scene graph
26. Step 2b: Load the model
27. Step 2c: Add it to the scene graph
28. Step 2: The result
29. Step 3: Set up physics is a mature C++ physics engine. Panda3d includes bindings to it, and it can be controlled through Python. To simulate physics, you need a B u l l e t W o r l d Keeps track of all the physics objects Computes new positions Handles collisions Physics objects consist of a shape (collisions) and a body (dynamics) Bullet
30. Step 3a: Create the physics world
31. Step 3b: Create collision geometry
32. Step 3c: Create a physical body
33. Step 3d: Add it to the physical world
34. Step 3: The result 0:05
35. Step 4: Link graphics and physics
36. Step 4a: Add the body to the scene graph
37. Step 4b: Link physics and graphics
38. Step 4: The result 0:07
39. Demo: “Glue blocks together to make the tower stable”
40. Demo: “Glue blocks together to make the tower stable” 0:23
41. Demo: “Bump the table”
42. Demo: “Angry blocks 3D”
43. Demo: “Angry blocks 3D” 0:31
44. Conclusion Question: How do people reason about the physical world? Hypothesis: "The physics engine in your head" Tools: Blender, Panda3D, and Bullet
45. Looking forward... -- , May 2012 (emphasis added) Angry Birds games had so far been played a total of 200,000 years, with 300 million minutes of playing time daily. TNW Blog Think how much we could learn if people participated in psychology experiments at that scale! With engaging and interactive physics experiments, it doesn't seem impossible.
46. Thanks! People Josh Tenenbaum Tom Griffiths Computational Cognitive Science Lab @ MIT Computational Cognitive Science Lab @ Berkeley Tools Python Panda3D, Blender, and Bullet IPython, NumPy, and the rest of the SciPy stack Slides created with IPython notebook and reveal.js
47. Links Slides and demo code: Project page: SceneSim library: PNAS: Scientific American: Fast Company: https://github.com/jhamrick/pycon-2014-talk http://web.mit.edu/~pbatt/www/PNAS2013/user.html http://scenesim.org/ http://www.pnas.org/content/110/45/18327.abstract http://www.scientificamerican.com/article/the-video-game-engine-in-your- head http://www.fastcodesign.com/3022236/evidence/how-our-brains-play- angry-birds-with-the-world
48. IPython Notebook