Inertial Properties in Robotic Manipulation: An Object-Level Framework
Authors: Khatib · Year: 1995 · Venue: International Journal of Robotics Research (IJRR)
Raw: md
Summary
This is a synthesis of Khatib’s operational-space formulation, extended to redundant serial (macro-/mini-) and parallel (multiarm) structures. The central technical contribution is the dynamically consistent generalized inverse of the Jacobian transpose, which decomposes joint torques into a part that produces operational-space end-effector forces and a part that drives only internal (null-space) motions without disturbing the end effector. It also develops object-level inertial models (effective mass/inertia, the “belted ellipsoid” visualization), the reduced-effective-inertia property of macro-/mini systems, and the augmented object model for the additive inertial properties of cooperating arms.
Key Claims
- For a redundant manipulator there exists a unique generalized inverse of
J^Tthat is dynamically consistent, i.e. null-space torques produce zero operational acceleration; it is the inertia-weighted inverseJ̄(q) = A^{-1} J^T (J A^{-1} J^T)^{-1}(Theorem 1). - This inverse yields a decomposition of joint torque
Γ = J^T F + [I − J^T J̄^T(q)] Γ_0, dynamically decoupling end-effector force control from internal-motion control (eq. 27). - Reduced effective inertia (Theorem 2 / Corollary 1): for a serial macro-/mini manipulator, the operational-space pseudo-kinetic-energy matrix satisfies
w^T Λ_o w ≤ w^T Λ_{m(o)} wfor all directionsw; the effective mass/inertia of the combined system is bounded above by that of the mini-manipulator alone, with equality for purely prismatic mechanisms. More generally, a redundant manipulator’s inertial properties are bounded above by those of the smallest distal subset of DOFs spanning operational space. - Augmented object (Theorem 3): the operational-space kinetic energy matrix of a multiarm/object system is the sum
Λ_⊕(x) = Λ_object(x) + Σ_i Λ_i(x); Coriolis/centrifugal and gravity terms are likewise additive. - Multiarm dynamic consistency (Theorem 4): each arm’s dynamically consistent inverse must account for the reflected load
Λ_⊕(q) − Λ_i(q_i)imposed by the object and the other arms. - The dynamically consistent framework also yields a singularity-handling strategy: a manipulator at a singular configuration is treated as redundant in the subspace orthogonal to the singular direction.
Method
Built on the operational-space equations of motion. Joint space: A(q) q̈ + b(q,q̇) + g(q) = Γ. Operational space: Λ(x) ẍ + μ(x,ẋ) + p(x) = F, with the operational inertia matrix obtained from the basic Jacobian J_0 via Λ = (J A^{-1} J^T)^{-1} (the “pseudo kinetic energy matrix” eq. 17 in the redundant case).
For redundancy the relationship Γ = J^T F becomes incomplete; the general form is Γ = J^T F + [I − J^T J̄^T] Γ_0 where the bracketed projector annihilates end-effector forces. Requiring null-space torques to produce no operational acceleration (Λ J A^{-1} (·) = 0) selects the unique inertia-weighted inverse J̄ = A^{-1} J^T Λ. Inertial properties are analyzed by separating translational (J_v) and rotational (J_ω) tasks, giving effective mass m_u along u and effective inertia I_u about u; the belted ellipsoid plots magnitudes (w ∝ v, |w| = v^T v) rather than the square-roots that ordinary inertia ellipsoids display.
The macro-/mini decomposition uses a block factorization of A (Golub & Van Loan) with A_22 the mini-manipulator block; Lemma 1 shows A_22 equals the mini-manipulator’s stand-alone kinetic-energy matrix, and nonnegative-definiteness of the Schur complement gives the reduced-inertia bound. Dextrous dynamic coordination then injects midrange-attraction torques (gradient of a joint potential) into the dynamically consistent null space so the mini-manipulator stays off its joint limits without perturbing the task.
Regime. The paper is fixed-base / fully-actuated; it is not a free-floating analysis (no momentum-conservation / generalized-Jacobian treatment of an uncontrolled base). Critically, Khatib explicitly maps the holonomic mobile manipulator and the free-FLYING space robot onto the macro-/mini paradigm: “This approach has been implemented for the coordination and control of a free-flying robotic system (Russakow and Khatib 1992)… the free-flying base, treated as a macro structure, and the manipulator, considered as the relatively lightweight mini structure.” This is the free-flying (actuated 6-DOF base) regime, treated as a redundant serial chain — directly relevant to our system rather than the free-floating literature.
Relevance to thesis
The dynamically consistent inverse is the canonical tool for splitting an actuated free-flying base + arm into a task (end-effector) controller plus a null-space behavior controller without cross-talk — exactly the decoupling our nominal guidance/control layer needs. The macro-/mini framing legitimizes treating the 6-DOF flying base as a slow macro stage and the arm as a fast mini stage, and the reduced-effective-inertia result quantifies what dynamic performance is reachable. The augmented-object additive property and reflected-load consistency matter if the manipulator grasps/handles an object. The singularity-as-local-redundancy idea connects to dynamic singularity handling.
Connections
Topics: operational_space_formulation · dynamically_consistent_inverse · null_space_projection · macro_mini_manipulation · augmented_object_model
Key Equations / Quotes
“A generalized inverse of J(q) satisfying the above constraint is said to be dynamically consistent (Khatib 1990).” — Theorem 1, p. 23. Result:
J̄(q) = A^{-1} J^T (J A^{-1} J^T)^{-1}.
Torque decomposition (eq. 27):
Γ = J^T F + [I − J^T J̄^T(q)] Γ_0, dividing torque into “joint torques corresponding to forces acting at the end effector (J^T F); and joint torques that only affect internal motions.” (p. 23)
Reduced inertial properties (Theorem 2, p. 28):
w^T Λ_o w ≤ w^T Λ_{m(o)} w— “The effective mass (inertia) … of a macro-/mini-manipulator system is smaller than or equal to the effective mass (inertia) associated with the mini-manipulator.”
Augmented object (Theorem 3, p. 31):
Λ_⊕(x) = Λ_object(x) + Σ_{i=1}^{N} Λ_i(x).
“This approach has been implemented for the coordination and control of a free-flying robotic system (Russakow and Khatib 1992) … the free-flying base, treated as a macro structure, and the manipulator, considered as the relatively lightweight mini structure.” (p. 29)
Open Questions
- The free-flying application is only cited (Russakow & Khatib 1992); how is base-actuation cost / fuel weighed against arm motion when both are actuated — does the “minimize instantaneous kinetic energy” criterion still give a sensible base/arm split?
- Equality of the reduced-inertia bound holds for prismatic mechanisms; what is the gap for a revolute arm on a flying base, and does it bound achievable bandwidth?
- The augmented-object reflected-load consistency (Theorem 4) assumes rigid grasp; how does it degrade under compliant or uncertain grasp — relevant once a risk layer is added.