Robot control
Active system identification
Definition
Active system identification chooses robot inputs or trajectories specifically to make unknown dynamics or physical parameters easier to estimate from the resulting measurements. It treats data collection as part of the identification problem rather than accepting only passively recorded motion.
Also known as: Active system ID
Updated
Design an informative experiment
Ordinary system identification fits a model to measured inputs and outputs. Active system identification also chooses the inputs. A robot might move selected joints, vary contact loads, or follow a planned trajectory so that mass, friction, compliance, or actuator effects leave distinguishable signatures in its measurements.
The experiment must excite the dynamics of interest. Repeating a nearly static motion may provide many samples but little information about inertia. A very fast motion may expose inertia while making friction and delay difficult to separate. The MIT system-identification notes show why the chosen model structure and loss determine which parameters the data can identify.
Useful data, not arbitrary exploration
An active design can optimize a measure of expected parameter information subject to motion, actuator, and safety constraints. Fisher information is one such measure. The 2026 Informationally Decoupled Trajectory Design paper studies trajectories intended to separate the effects of different simulator parameters. Its authors report simulation experiments on several robots and a physical K1 humanoid experiment. Those are results for that method and setup, not evidence that every informative trajectory transfers safely to hardware.
This differs from trajectory optimization for task performance. A task trajectory is chosen to reach or manipulate something; an identification trajectory is chosen to reduce uncertainty about a model. One motion can serve both goals, but neither objective implies the other.
Identifiability and safety remain limits
Some parameters have indistinguishable effects under the available sensors and motions. Increasing excitation does not resolve a structural ambiguity, and correlated noise can make an information calculation optimistic. Unmodelled flexibility, backlash, temperature, contact changes, and controller dynamics can also be absorbed into the wrong parameter.
Exciting a large robot adds practical limits. Joint travel, balance, collision clearance, thermal load, and human separation must constrain the experiment. The fitted model then needs validation on motions that were not used for fitting. A close replay of the identification trajectory alone does not establish useful prediction elsewhere.
Sources
Related terms
System identification
System identification estimates a model of a physical system from measured inputs and outputs. In robotics, it can recover parameters such as inertia and friction or learn a more general model of how actions change the system state.
Sim-to-real transfer
Sim-to-real transfer applies a model, policy, or behavior developed in simulation to a physical system. Its central challenge is the difference between the simulated environment and the robot, sensors, and interactions encountered in reality.
Trajectory optimization
Trajectory optimization finds a time-varying motion, and often control inputs, that minimizes an objective while satisfying specified constraints. Robot applications can include geometric, kinematic, and dynamic constraints.
Sensor fusion
Sensor fusion combines information from multiple sensors or estimation sources to produce a shared estimate. The combination must account for coordinate frames, timing, uncertainty, and dependence between inputs.