Measure joint or wheel motion.
Encoders provide local position or velocity information.
Side 60
Robotics
A study of machines that sense, estimate, decide and act in the physical world. Robotics integrates mechanics, electronics, control, computation and uncertainty into closed-loop autonomous behavior.
sense→estimate→plan→control→act
06robotic layers
05motion questions
05autonomy problems
60Side
A robot only knows the world through sensors.
Measurements are partial, noisy and delayed, so perception is an inference problem.
Measure acceleration and rotation.
Useful for short-term motion estimation but accumulates drift.
Rich visual observation.
Images support recognition, geometry and tracking but depend on lighting and viewpoint.
Measure distance by light.
Produces geometric range data for mapping and obstacle detection.
Measure interaction.
Force/torque sensors help robots manipulate and contact the environment safely.
Detect nearby objects.
Simple sensors can provide robust local safety cues.
State estimation combines imperfect evidence.
Robots must infer position, velocity and environment state from noisy measurements and motion models.
What did we believe before?
Previous state.
State estimates propagate forward using a motion model.
Where should the robot be now?
Control + dynamics.
Prediction introduces uncertainty from imperfect models and actuation.
What do sensors observe?
Noisy evidence.
Measurements provide corrections but contain their own uncertainty.
How should prediction and measurement combine?
Weight by uncertainty.
Filtering balances trust in model and sensors.
What is the new state estimate?
Belief, not certainty.
The output is an estimate with residual uncertainty.
Kinematics maps joints to motion.
Robots need geometric models that connect actuator coordinates to position and orientation in space.
Frame
Define coordinate systems.
Transformations relate robot, sensor, tool and world coordinates.
Forward
Joint values → end-effector pose.
Forward kinematics computes where the mechanism reaches.
Inverse
Desired pose → joint values.
Inverse kinematics may have multiple, singular or no valid solutions.
Jacobian
Relate joint rates to task-space velocity.
The Jacobian reveals local mobility and singular configurations.
Workspace
Which poses are reachable?
Geometry and joint limits define the robot’s reachable space.
Planning turns goals into collision-free actions.
The robot must choose paths through a space constrained by obstacles, dynamics and task requirements.
Represent possible robot states.
Obstacles in the world become forbidden regions in configuration space.
Search discretized possibilities.
A* and related algorithms find paths using cost and heuristic guidance.
Explore large continuous spaces.
RRT and PRM methods build feasible paths without exhaustively discretizing everything.
Add time and dynamics.
A geometric path becomes executable only after velocity and acceleration constraints are respected.
Respond to a changing environment.
Autonomous robots must revise plans when obstacles or goals move.
Choose among feasible paths.
Distance, time, energy, safety margin and smoothness can all shape the objective.
Planning says what should happen; control makes the body follow.
Feedback corrects deviation between desired trajectory and measured state.
| Layer | Goal | Typical issue |
|---|---|---|
| Position | Reach commanded pose | Overshoot / steady-state error |
| Velocity | Track commanded speed | Noise / lag |
| Force | Regulate contact | Instability at interaction |
| Impedance | Shape force-motion relation | Safe compliant behavior |
| Model predictive | Optimize future constrained motion | Computation and model accuracy |
Autonomy closes the entire loop.
A useful autonomous robot must continue sensing and revising its state, plan and action as reality diverges from expectation.
Localize
Estimate where the robot is.
Map
Represent relevant structure in the environment.
Interpret
Identify objects, obstacles and task-relevant states.
Decide
Select behavior consistent with goals and constraints.
Execute
Control actuators while observing the result.
Fail safe
Detect uncertainty or fault conditions that require stopping, fallback or human intervention.
Probabilistic RoboticsThrun, Burgard & Fox · estimation and mapping
Modern RoboticsLynch & Park · mechanics, planning and control
Introduction to Autonomous RobotsCorrell et al. · broad robotics foundation
Planning AlgorithmsSteven LaValle · motion planning