model predictive control
A control method that casts out many possible future paths per second and prunes them based on predicted outcomes such as collision likelihood.
As defined in 30 | Derek Leben on Ethics for Robots and Artificial Intelligences
Other definitions
A classical optimal control technique that uses learned or known models of system dynamics to predict future states under different action sequences, then optimizes actions in real-time to minimize a cost function. Has existed since the 1960s.
The Shape of AI to Come! Yann LeCun at AI Action Summit 2025
Where it comes up · 5
- Marc Raibert4 claims
- The Shape of AI to Come! Yann LeCun at AI Action Summit 20252 claims
- 30 | Derek Leben on Ethics for Robots and Artificial Intelligences1 claim
- Yann LeCun: Special Lecture on AI and World Models1 claim
- Inference, Diffusion, World Models, and More | YC Paper Club1 claim
Most important claims · 9 of 9Search all claims for “model predictive control” →
- Model predictive control (MPC) can adapt to novel reward functions at test time because it factors the control problem into a dynamics model and an action selector mechanism, allowing the planner to optimize for different objectives without retraining the dynamics model.Inference, Diffusion, World Models, and More | YC Paper Club
- Reference to 'test time computation' and related ideas as recent innovations is misleading: this is classical optimal control under the name Model Predictive Control (MPC), which has existed since the early 1960s; variants that learned the dynamics model emerged in the 1970s from France in work like ECCOM, and the concepts are well-established in optimal control literature.The Shape of AI to Come! Yann LeCun at AI Action Summit 2025
- Model Predictive Control and world models are not new concepts but rather established techniques from the control theory community dating back to the 1970s and earlier, so the innovation is applying learned models to replace hand-written differential equations.Yann LeCun: Special Lecture on AI and World Models
- Model Predictive Control (MPC) is a classical control method from ~1960s that plans through optimization using forward models; more recent versions (1970s, from France, called EDCOM) learn the models, making test-time computation via optimization well-established and not new despite current enthusiasm about 'test-time compute'The Shape of AI to Come! Yann LeCun at AI Action Summit 2025
- Natural robot movement comes from taking a dynamic approach that predicts the future evolution of motion and uses that prediction when sending signals, as opposed to purely reactive control that only looks backward at the current state and tries to adjust for it.Marc Raibert
- Because neither human drivers nor AI can predict outcomes with certainty, self-driving cars will constantly make decisions between higher-risk and lower-risk actions even when actual collisions are rare, so likelihood-based path pruning (as in model predictive control) must be supplemented by weighting the severity of likely collisions—a likely collision with a pedestrian is worse than one with a tree.30 | Derek Leben on Ethics for Robots and Artificial Intelligences
- Spot uses a limited-horizon control approach that only predicts about a second to a second and a half ahead, recalculating roughly every tenth of a second, while maneuvers like somersaults require planning much further ahead to coordinate momentum and rotation for landing.Marc Raibert
- Despite huge optimism about machine learning since ChatGPT, doing physical things with physical robots differs from language because the tokens don't exist and pixel values aren't like words; the most impressive robot performances, such as Atlas, are still mostly traditional model predictive control rather than learning.Marc Raibert
- For athletic intelligence, the future will be a mating of traditional model predictive control and machine learning, yielding the best of both worlds.Marc Raibert