Building and Assessing Artificial Intelligence
In the past decade, there have been impressive strides made in developing AI and robotics systems. Yet despite great accomplishments, there is no robot that is able to interact with the physical world as quickly, robustly, and flexibly as humans can. I use my research reverse-engineering human physical common sense to bridge this machine-to-human gap by designing AI systems that learn and reason in human-like ways, integrating physical reasoning into robots, and assessing the capabilities of artificial systems to perform human-level reasoning.
Related publications:
- Modeling expectation violation in intuitive physics with coarse probabilistic object representations
- Physion: Evaluating physical prediction from vision in humans and machines
- Physion++: Evaluating physical scene understanding that requires online inference of different physical properties
- Probabilistic simulation supports generalizable intuitive physics
- Data-efficient learning for complex and real-time physical problem solving using augmented simulation
- Are deep neural networks SMARTer than second graders?
- H-SAUR: Hypothesize, Simulate, Act, Update, and Repeat for understanding object articulations from interactions
- Unsupervised discovery of 3D physical objects from video
- End-to-end differentiable physics for learning and control
- Differentiable physics and stable modes for tool-use and manipulation planning
