Modern Robot Learning Text Only
Foundation model A large pretrained model that can be adapted to many tasks.
Modern Robot Learning Text Only
Pretraining Training a model on a broad dataset before adapting it to a specific task.
Modern Robot Learning Text Only
Fine-tuning Taking a pretrained model and adapting it to a specific robot or task.
Modern Robot Learning Text Only
Generalist robot model A model trained to perform many tasks instead of just one.
Modern Robot Learning Text Only
Vision-Language Model (VLM) A model that understands both images and text.
Modern Robot Learning Text Only
Vision-Language-Action model (VLA) A model that takes images and language as input and outputs robot actions.
Modern Robot Learning Text Only
Language-conditioned control Robot control that depends on a natural-language instruction.
Modern Robot Learning Text Only
World model A model that predicts how the world will change after actions.
Modern Robot Learning Text Only
Multimodal Using more than one type of input, like vision, language, touch, or proprioception.
Modern Robot Learning Text Only
Open-vocabulary The ability to understand names or categories beyond a fixed label set.
Modern Robot Learning Text Only
Zero-shot Doing a new task without task-specific training.
Modern Robot Learning Text Only
Few-shot Learning a new task from only a small number of examples.
Modern Robot Learning Text Only
Generalization The robot’s ability to work in new situations it has not seen before.
Modern Robot Learning Text Only
Transfer learning Using knowledge from one task, domain, or robot to help with another.
Modern Robot Learning Text Only
Cross-embodiment transfer Transferring knowledge across different robot bodies.
Modern Robot Learning Text Only
Domain adaptation Adapting a model to work in a different environment or data distribution.
Modern Robot Learning Text Only
Diffusion policy A robot policy that generates actions using diffusion-model techniques.
Modern Robot Learning Text Only
Transformer policy A robot policy built with a transformer architecture.
Modern Robot Learning Text Only
Sequence modeling Treating robot behavior as a sequence problem over time.
Modern Robot Learning Text Only
Autoregressive policy A policy that predicts actions one step at a time in sequence.
Modern Robot Learning Text Only
Action chunking Predicting several future actions at once instead of one action at a time.
Modern Robot Learning Text Only
Chunk size How many future actions are predicted together in one chunk.
Modern Robot Learning Text Only
Action horizon How far into the future the policy predicts or plans.
Modern Robot Learning Text Only
Temporal abstraction Reasoning over longer time spans instead of only single low-level steps.
Modern Robot Learning Text Only
Skill A reusable behavior like grasp, push, place, or open drawer.
Modern Robot Learning Text Only
Primitive / action primitive A simple reusable low-level movement or control building block.
Modern Robot Learning Text Only
Skill library A collection of reusable skills that can be composed into larger tasks.
Modern Robot Learning Text Only
Hierarchical policy A policy with multiple levels, where higher levels choose goals or skills and lower levels execute them.
Modern Robot Learning Text Only
Long-horizon task A task requiring many coordinated steps, memory, or replanning.
Modern Robot Learning Text Only
Multi-stage task A task made of several subgoals or phases.
Modern Robot Learning Text Only
Task decomposition Breaking a large task into smaller subproblems.
Modern Robot Learning Text Only
Recovery behavior Behavior that helps the robot get back on track after a mistake.
Modern Robot Learning Text Only
Failure recovery A system’s ability to detect and recover from errors.
Modern Robot Learning Text Only
Robustness How well a robot keeps working despite noise, disturbances, or variation.
Modern Robot Learning Text Only
Affordance What actions an object allows, such as a handle being pullable or a button being pressable.