Galilean Symmetry in Robotics
summary
The gist
Galilean symmetry is presented as a roboticstailored exposition of Galilean symmetry that leverages the community’s familiarity with and understanding of rigid-body transformations and pose…
In short
The paper explores Galilean symmetry in robotics by using a matrix Lie group to describe two pose representations: Galilean frames based on inertial velocity and extended poses based on coordinate velocity. It demonstrates how this framework is useful for solving problems in inertial navigation, manipulator kinematics, and sensor data fusion under temporal uncertainty.
Key concepts
- Galilean Group (Gal(3))
- This is a 10-dimensional Lie group that describes Galilean transformations in space and time. It includes translations, spatial rotations, and velocity boosts. It forms the mathematical foundation for relating different inertial reference frames in classical mechanics.
- Galilean Frames vs. Extended Poses
- The paper distinguishes between two ways to represent a robot's pose: Galilean frames use inertial velocity as the state definition, while extended poses use coordinate velocity. This distinction is crucial because it allows for different kinematic formulations depending on the application.
- Temporal Uncertainty Modeling
- Time is explicitly incorporated into the Galilean framework, allowing temporal uncertainty to be modeled. This uncertainty is encoded using a concentrated Gaussian distribution around an estimated state, which helps in robust state estimation when measurements are noisy.
Terminology used across episodes
This episode discusses
The paper
Galilean Symmetry in Robotics · Read on arXiv
Robert Mahony, Jonathan Kelly, Stephan Weiss
Australian National University · University of Toronto · University of Klagenfurt
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "Galilean Symmetry in Robotics".
Paper summary: Rosa: So we're looking at this paper called "Galilean Symmetry in Robotics". The main idea here seems to be taking Galilean symmetry, which is really the natural symmetry for inertial motion in Newtonian physics, and making it relevant for robotics.
Dev: Right. It seems they're not just treating it as a stepping stone to relativity but framing it directly for robotics people who are already familiar with rigid-body transformations and pose representations.
Rosa: That’s the core of what they claim is important here, that this approach contrasts with how physics literature usually introduces Galilean symmetry. They want to show how the Galilean matrix Lie group can be used to describe two different ways of representing a robot's pose: Galilean frames and extended poses.
Taro: So you have these two representations, one using inertial velocity and the other using coordinate velocity, right? That distinction seems key for setting up the whole framework.
Dev: Exactly. The paper points out that they can use this Galilean matrix Lie group to handle both of those pose representations simultaneously. This opens up a way to think about how motion is modeled in different contexts within robotics applications.
Rosa: They go on to give three examples where applying this Galilean matrix Lie group algebra seems pretty straightforward and actually yields some pretty useful insights for practical problems. I'm looking at inertial navigation above the rotating Earth, manipulator kinematics, and sensor data fusion under temporal uncertainty.
Taro: That sounds interesting, especially the part about inertial navigation in a frame that’s already moving due to rotation. What does that imply for how we model things when the reference frame itself is turning?
Dev: It implies they have a concrete way to handle that motion within the symmetry structure. They build kinematics around Galilean frames where you explicitly include that inertial velocity and event position in the state definition, like they describe it on page two of this paper <ref:2510.10468#pg2>.
Rosa: And then there's the other side with extended poses, which uses coordinate velocity instead of inertial velocity in its definition. This is where things get applied to things like manipulator kinematics.
Taro: When you talk about manipulator kinematics, how does extending the Denavit-Hartenberg transformation work when you're using coordinate velocity instead of just inertial velocity?
Dev: They extend the Generalized Denavit-Hartenberg transformation to include both joint angular and linear velocities. The resulting forward kinematics for a serial manipulator ends up looking like a product of these transformed matrices, which is written as zero K n equal to zero K one times zero K two times... n minus one Kn.
Rosa: And they use this structure to model that motion. But they also tackle temporal uncertainty by naturally incorporating time into the Galilean framework. The robot's state is represented by a Galilean frame F k, which includes the rotation matrix, the inertial velocity vector, and the position and time of an event.
Paper summary: Taro: So they use a concentrated Gaussian structure to encode that temporal uncertainty in this frame representation? How does that translate into something tangible for state estimation?
Dev: They use a concentrated Gaussian structure where the uncertainty in information state is encoded by a parameter eta k which follows a normal distribution with zero mean and covariance Q k. This setup allows them to model the measurement events as being affected by noise, like y tau i equal F i y zero plus mu i, where the noise mu i is also normally distributed.
Rosa: The paper then moves into Bayesian fusion by transforming this uncertainty problem into error coordinates around the estimated state F hat k. They define an error variable Ek which is basically the difference between the true frame and the estimated frame, and that error variable follows a normal distribution with covariance Q k.
Taro: So they are optimizing the log-likelihood of a certain expression to figure out what the mean estimate of this information state would be in those local coordinates. What does that actually give us in terms of the final fused state?
Dev: The result is a fused information state, which they describe as a concentrated Gaussian distribution p k equals normal with mean F star k and covariance Q star k. This means they have a probabilistic description of the robot's pose based on measurements.
Rosa: It seems like this whole approach—using the Galilean matrix Lie group to bridge inertial and coordinate velocity representations, handling temporal uncertainty explicitly, and then fusing that data in error coordinates—is designed to be very robust for real-world robotics problems.
Taro: So what's the big picture here? Why focus on Galilean symmetry instead of just using standard rigid-body tools? What’s the actual value added by this structure?
Dev: The value is in providing a broader geometric foundation that explicitly incorporates both time and motion into the symmetry itself. It gives a richer structure to analyze how motion happens, which is what they hope inspires further work in the robotics community.
Rosa: So, to wrap up, "Galilean Symmetry in Robotics" presents this robotic-tailored way of looking at Galilean symmetry by contrasting inertial and coordinate velocity representations using the Galilean matrix Lie group. It shows how this structure applies cleanly to navigation above rotating Earth, manipulator kinematics with extended poses, and sensor fusion under temporal uncertainty.
Dev: And what they conclude is that this symmetry provides a broader geometric foundation for robotics that explicitly includes time and motion in a way that connects nicely to classical physics concepts while being tailored for the robotics community.
Conclusion: Rosa: So we're wrapping up this look at "Galilean Symmetry in Robotics." This paper is basically taking the old physics symmetry of Galilean transformations and applying it directly to how robots move and are described in real systems.
Dev: Yeah, they use this Galilean matrix Lie group to connect two different ways of describing a robot's pose—one using inertial velocity and another using coordinate velocity. It’s a way to build a geometry that handles both kinds of motion at once.
Rosa: And they show this framework works for things like navigating robots above the Earth while it's spinning, or figuring out how robotic arms move when you use those extended pose representations instead of just inertial ones.
Dev: The big takeaway is that this symmetry gives us a more complete geometric foundation for robotics because it explicitly includes both time and motion in the way we define the system. It’s not just a tool for one specific problem; it sets up a broader way to think about spatial awareness in these systems.
Rosa: Exactly, so when you look at where they started with inertial navigation or sensor fusion, this symmetry provides that deeper structure to analyze those problems. It’s about making sure the math you use actually reflects the physics of motion better.
Dev: It’s a nice way to connect classical concepts—like Galilean transformations—to modern robotics challenges, but you have to keep in mind that it's tailored specifically for how we model these state representations. We still have to worry about the real-time loop rates and what happens when the world misbehaves.
Rosa: Right, so even with this elegant symmetry structure, you still need to check if the latency is low enough for a fast robot to use it in practice when things get messy. That’s where we go next.
More episodes
- 2608.13438-ContactGuard: Pre-Contact Execution Monitoring with Action-Conditioned Latent World Models
- 2610.34593-Future Prediction as Training-Only Video Supervision for Pretrained VLA Policies
- 2610.35385-EmbodiedLab: Beyond Manipulation, Benchmarking Embodied Agents That Act to Know
- 2610.36378-Data-Driven Hamiltonian for Neural Optimal Control
- 2610.37275-A Heuristic for Scheduling Autonomous Vehicles at Intersections
- 2610.37288-Dynamic Coordination of Autonomous Vehicles at Intersections
- 2610.37632-Real-Time Coordination of Autonomous Vehicles at Intersections
- 2611.13779-Softness is something you also need! Why compliance helps control: information metrics of a passive walker with compliant ankle
- 2611.13411-Residual-Corrected Physics-Informed Neural Network KKL Observers with Input-to-State Stability Guarantees
- 2611.14488-Is Your Robot's Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies