A catalog of inverse-kinematics planners for underactuated systems on matrix Lie groups

Sonia Martínez, Jorge Cortés, Francesco Bullo

TL;DR

This paper provides a "catalog" of closed-form algebraic solutions for the motion planning of underactuated systems—robots with fewer actuators than degrees of freedom—evolving on common geometric spaces like planes and 3D rotations. By decomposing complex trajectories into a sequence of simple "motion primitives," the authors transform a difficult nonlinear control problem into a straightforward algebraic one.

What Problem Does This Paper Address?

In robotics, we often deal with underactuated systems: machines that cannot move in every direction simultaneously. Think of a car (which cannot move sideways without maneuvering) or a satellite with only two thrusters. Mathematically, these systems are often modeled on Lie groups (like $SE(2)$ for planar motion or $SO(3)$ for 3D rotations).

Traditionally, planning a path for these robots requires complex nonlinear optimization or iterative numerical methods, which are computationally expensive and prone to getting stuck in local minima. The "holy grail" is a closed-form solution: a direct formula where you plug in your start and end points and get the exact control sequence needed. Before this paper, such solutions were scattered across the literature or non-existent for many common system configurations.

Methods

The authors utilize a "kinematic reduction" approach, focusing on driftless systems where the robot only moves when a control input is applied.

The core technique involves:

  1. Transcription: They represent motion as a sequence of $k$ "motion primitives" (flows along specific vector fields). The goal is to solve the equation: $g_f = \exp(t_1 V_{i1}) \dots \exp(t_k V_{ik})$, where $g_f$ is the target and $t_i$ are the "coasting times" spent on each primitive.
  2. Controllability Analysis: Using Lie algebra, they classify systems based on whether they can reach any configuration in their space.
  3. Inverse-Kinematics (IK) Maps: For each classified system, they derive explicit algebraic formulas (IK maps) to calculate the required coasting times.
Figure 1: Motion primitives for a planar rigid body on $SE(2)$.

Key Insights and Ideas

  • The Power of Geometry: By treating the robot's state space as a matrix group, the authors can use the elegant tools of Lie theory to solve control problems that look messy in standard coordinates.
  • Classification is Key: The paper shows that not all underactuated systems on the same group are equal. For example, on $SE(2)$, they distinguish between "S1" and "S2" systems based on their controllability properties, requiring different planning strategies.
  • The "Switching" Trade-off: While they aim for the minimum number of switches, they discover a "negative result" for certain 4D systems ($SE(2) \times \mathbb{R}$). In these cases, a 4-primitive (switch-optimal) solution is impossible near the starting point, necessitating a 5-primitive sequence instead.
Figure 5: A planned trajectory for an underactuated system on $SE(2) \times \mathbb{R}$.

Key Results

  • Catalog of Solutions: The paper provides explicit, ready-to-implement formulas for systems on $SE(2)$ (planar), $SO(3)$ (rotational), and $SE(2) \times \mathbb{R}$ (planar + one extra dimension, like a snakeboard).
  • Controllability Lemmas: They provide rigorous conditions to check if a given set of actuators is sufficient to move the robot anywhere.
  • Closed-Form IK: For $SO(3)$, they relate the motion planning problem to Euler angles, providing a geometric intuition for 3D rotation planning.

Limitations and Future Work

  • Driftless Assumption: The models are kinematic, meaning they ignore momentum and inertia. This works well for slow-moving robots but not for high-speed dynamic maneuvers.
  • Obstacle Avoidance: The catalog provides "feasible" paths, but it does not inherently account for avoiding walls or obstacles in the environment.
  • Path Optimality: While the planners are "switch-optimal" (minimum number of stops/starts), they are not necessarily "distance-optimal" (the shortest possible path).

Summary updated

Abstract

This paper presents motion planning algorithms for underactuated systems evolving on rigid rotation and displacement groups. Motion planning is transcribed into (low-dimensional) combinatorial selection and inverse-kinematics problems. We present a catalog of solutions for all underactuated systems on SE(2), SO(3) and SE(2) /spl times/ /spl Ropf/ classified according to their controllability properties.
BibTeX
@inproceedings{martinez2003catalog,
  author = {Sonia Martínez and Jorge Cortés and Francesco Bullo},
  title = {{A catalog of inverse-kinematics planners for underactuated systems on matrix Lie groups}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2003},
  doi = {10.1109/iros.2003.1250699},
}

Loading related content...

Figure