Glossary

Each term links to the lesson that teaches it.

Term	In short	Lesson
ACT (Action Chunking with Transformers)	Imitation policy that predicts a chunk of future actions with a CVAE-trained transformer	31
Action chunking	Predicting several future actions at once and executing them before re-querying	31
Antipodal grasp	Two contacts whose normals point at each other, inside both friction cones	16
Armature	Reflected rotor inertia added to a joint; stabilizes stiff position control	05
Behavior cloning (BC)	Supervised learning of observation → action from demonstrations	28
body / geom	In MJCF: a moving frame / a collision and mass shape attached to it	01
Collocation	Trajectory optimization with states and controls as decision variables tied by dynamics constraints	23
Configuration space (C-space)	The space of all joint configurations; obstacles become forbidden regions in it	22
Contact normal	The direction the engine uses to push two bodies apart	02
Creep	Slow steady sliding below the slip threshold caused by soft contact	04
DAgger	Dataset aggregation: label the states the learner visits with an expert, retrain	29
Damped least squares (DLS)	IK update $(J^\top J + \lambda^2 I)^{-1} J^\top e$ that stays bounded near singularities	10
Decimation	Number of physics steps between two policy decisions	34
Domain randomization	Training across randomized physics so the policy covers the real system	39
Flow matching	Generative model that learns a velocity field carrying noise to data; used for action generation	30
Forward kinematics (FK)	Joint angles → pose of a frame on the robot	08
Friction cone	Set of contact forces friction can sustain without slipping	04
Gimbal lock	Loss of one rotational degree of freedom in an Euler-angle parameterization	06
Gravity feedforward	Adding the model’s gravity torque to the controller output to remove sag	12
Homogeneous transform	4×4 matrix combining rotation and translation; composes frame changes	07
Homography	3×3 projective map between an image and a plane	20
Impedance control	Making the end effector behave like a spring–damper around a target	14
Intrinsics	Camera matrix $K$: focal lengths and principal point	19
Inverse kinematics (IK)	Finding joint angles that reach a desired pose	10
Jacobian	Matrix mapping joint velocities to end-effector velocity	09
Kalman filter	Recursive optimal state estimator for linear-Gaussian systems	18
Kp / Kd	Proportional and derivative gains of a PD controller	12
LQR	Optimal linear state feedback for linear dynamics and quadratic cost	24
Manipulability	Measure of how well an arm can move in all directions at a configuration	10
MJCF	MuJoCo’s XML format for describing a world	01
MPC	Re-solving a finite-horizon optimal control problem at every step	24
Operational-space control	Task-space control using the end-effector inertia $\Lambda = (J M^{-1} J^\top)^{-1}$	14
Overshoot	How far the response goes past its final value	12
PPO	Policy-gradient algorithm with a clipped surrogate objective	33
Quaternion	Unit 4-vector representing a rotation; MuJoCo stores it w-first	06
Replay	Sending recorded actions again without looking at the world (open loop)	26
Reward shaping	Adding terms to the reward to guide learning; potential-based shaping keeps the optimum	35
RRT / RRT-Connect	Sampling-based motion planners that grow random trees in C-space	22
Semi-implicit Euler	Integrator that updates velocity first, then position with the new velocity	02
Singularity	Configuration where the Jacobian loses rank	10
solref / solimp	MuJoCo contact stiffness/damping and impedance parameters	03
System identification	Fitting model parameters to measured data	38
Temporal ensembling	Averaging overlapping action chunks with exponential weights	31
timestep (dt)	How far the physics engine advances time in one step	02
Tunneling	A fast body crossing a thin obstacle between two steps	02
URDF	ROS robot description format	01
VLA	Vision-language-action model: images + language + state → actions	41
Watchdog	Timer that stops the robot when commands stop arriving	15
Wilson interval	Confidence interval for a success rate from k successes in n trials	25
World model	Learned model that predicts the next state from state and action	43