Model-Predictive Spill-Free Emergency Stopping for Liquid-Manipulating Robots
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "Model-Predictive Spill-Free Emergency Stopping for Liquid-Manipulating Robots".
Dev: Here's the gist of the paper. Manipulating open liquid containers is challenging because liquids are highly sensitive to vessel accelerations and jerks This letter presents an emergency stop system for robots manipulating liquids in open containers We formulate emergency stopping as an optimal control problem and solve it in a model predictive control framework to generate time-optimal, spill-free stopping trajectories The proposed method works as a plug-and-play safety layer on top of existing slosh-free motion planning methods, enabling immediate reaction to detected hazards while accounting for nonlinear liquid dynamics We demonstrate, through simulation and on a 7-DoF Franka Emika Panda robot, that the proposed approach achieves fast emergency stopping without spilling Problem Formulation and Approach We address the problem of bringing a robotic manipulator arm carrying a liquid container to rest as quickly as possible, without spilling the liquid, after an emergency stop is triggered This means minimizing the cumulative sum of the Cartesian velocities while obeying no-spill constraints for the allowed liquid-surface inclination The proposed method uses a standard model predictive control approach, in which we repeatedly solve optimal control problems with a receding time horizon We use a well-established spherical pendulum model of liquid slosh for its computational efficiency and reasonable modeling accuracy The problem is formulated as a Quadratic Program (QP) to minimize the cumulative sum of the end-effector’s (container’s) Cartesian velocities, while respecting the robot’s kinematic limits and no-spill constraints Modeling Liquid Dynamics Mechanical models, such as the spherical pendulum model or mass-spring model, are well-established methods for analyzing and predicting liquid slosh The spherical pendulum model's state is parametrized by two tilt angles (θ, ϕ) about the horizontal axes in the world frame and the pendulum rod length (l) We maintain bilinear coupling terms between z¨p and θ, ϕ so that we end up with bilinear small-angle dynamics To use the pendulum model, the rod length l must be estimated based on the container’s shape and size using natural frequencies of slosh The state space of our model comprises the pendulum angles and angular velocities, the end-effector’s 6-DoF Cartesian velocities, and the end-effector’s 3-DoF Cartesian orientation, totaling 13 state variables We have six control variables, namely the container’s Cartesian accelerations The optimization is run in a loop in receding horizon fashion with 20 Hz frequency, while a high-frequency real-time control loop reads the solution from shared memory and always communicates the first control action to the robot Until the emergency stop is initiated, we assume that the liquid surface is perpendicular to the container’s vertical axis After the emergency stop is initiated, we switch to integrating the pendulum state using the realized Cartesian accelerations Experimental Validation and Results The real-world experiments aimed to validate the proposed method with real hardware and liquid We compared the proposed method with a baseline that does not include slosh modeling The results show that the proposed method kept the maximum slosh essentially within the given limit of 5 degrees, only violating it by practically negligible 0.3 degrees On the other hand, the baseline method significantly exceeded this limit and produced maximum slosh of 22.
Title and authors: Rosa: So, what they actually propose is a way to integrate this emergency stop system into existing robot navigation plans so that it can react immediately when a hazard is detected. It’s not just about stopping; it’s about having this real-time safety feature that accounts for those complex, nonlinear liquid dynamics.
Dev: They use a spherical pendulum model for the liquid slosh because they said it's computationally efficient enough for real-time use, even though the fluid dynamics are really messy. They then formulate the whole thing as a quadratic program or QP to minimize how fast the container moves while respecting those no-spill constraints.
Taro: I wonder about that linearization effort they mentioned; they said they have to linearize some of those bilinear terms around a nominal trajectory just to make it run in real time without losing predictive power. That’s a huge hurdle for any real system, right?
Rosa: It is, Taro. It means the model has to be very smart about how it calculates things at every single time step in the optimization loop. They use this receding horizon approach with a frequency of twenty hertz to keep things moving fast.
Dev: And they also have this separate layer where an AI system solves another small quadratic program using something called Resolved Acceleration Control or RAC to convert those task space commands into actual joint space commands for the robot arms. That’s how you actually get the motors to move correctly.
The paper's summary: Rosa: The paper highlights a few key improvements over just having a standard slosh-free planner that suddenly gets an emergency override. They focus on making sure this new method is robust and fast enough for actual use outside of the lab.
Dev: They show through simulation and on this seven-degree of freedom Franka Emika Panda robot that they can achieve fast emergency stopping without spilling, which is a pretty strong validation point for their whole approach.
Taro: I'm interested in what they found when they compared their method against a baseline that didn't include any slosh modeling at all. They reported that the proposed method kept the maximum slosh essentially within a limit of five degrees, only violating it by practically negligible zero point three degrees.
Rosa: That comparison with the baseline is really telling, because in that baseline case, they saw maximum slosh of twenty two degrees and significantly exceeded the no-spill limit. That shows how much that liquid modeling actually matters when you need to stop suddenly.
Dev: The improvement they're pushing is moving from just planning motion to having this immediate reaction capability based on detected hazards while still accounting for the nonlinear dynamics of the liquid. It’s about seamless transition between planned motion and emergency action.
The paper's improvements: Rosa: So, to wrap up, this paper, "Model-Predictive Spill-Free Emergency Stopping for Liquid-Manipulating Robots," shows a way to treat emergency stopping not just as a hard stop but as an optimization problem that balances speed and safety while handling the liquid's weird physics.
Dev: Essentially, it gives robots the ability to stop quickly and safely under unexpected hazards by using model predictive control on top of existing motion planning methods, which is a very practical safety feature for deployment.
Taro: For me, what this means is that we can start thinking about how these systems recover gracefully when things go wrong in the real world, because they aren't just stopping; they are finding the best way to stop without making a mess.
Rosa: Right. So we’ve seen how they use the spherical pendulum model and MPC to create time-optimal, spill-free paths, and it seems like a solid way forward for keeping robots safe around liquids.
Dev: It's definitely a step toward building more capable safety layers on top of current slosh-free motion planning methods.
Taro: We’ll see how this plays out when we test these kinds of real-time control loops against actual unexpected disturbances in dynamic environments.
Conclusion: Rosa: So, to wrap up, this paper "Model-Predictive Spill-Free Emergency Stopping for Liquid-Manipulating Robots" shows how you can treat emergency stopping as a math problem that balances speed and safety while handling the liquid's complex physics.
Dev: It’s about formulating it as an optimal control problem solved in a model predictive control framework, which means we get those time-optimal, spill-free stopping trajectories.
Taro: I just want to emphasize the plug-and-play aspect; this layer works on top of existing slosh-free motion planning so you can react immediately when a hazard is detected without rewriting your whole planning infrastructure.
Rosa: Exactly. And they used a standard spherical pendulum model for the liquid slosh because it's efficient enough to run in real time, even though the fluid dynamics are nonlinear and messy.
Dev: They have this receding horizon approach running at twenty hertz frequency, and they linearize those problematic bilinear terms around a nominal trajectory just to make sure it runs fast without losing predictive power.
Taro: I did wonder about that linearization effort; it means the model has to be really smart about how it calculates things every single time step in that optimization loop.
Rosa: It’s a big hurdle for any real system, Taro, but they show through simulation and on this Franka Panda robot that they can achieve fast emergency stopping without spilling.
Dev: In the real-world test with actual liquid, they compared it to a baseline that didn't include slosh modeling at all, and their method kept the maximum slosh pretty much within a limit of five degrees.
Taro: That comparison is really telling; in that baseline case, they saw maximum slosh of twenty two degrees and it significantly exceeded the no-spill limit.
Rosa: So what this means for us is that we can build this kind of real-time safety layer on top of our motion planning methods to handle those sudden liquid hazards without a complete overhaul.
Dev: It gives us a mathematically optimized path to rest under emergency conditions, balancing the need for speed against the absolute requirement of spill prevention.
Taro: It’s definitely a step toward building more capable systems that can handle unexpected disturbances in dynamic environments gracefully and recover quickly.
Rosa: Yeah, it's about moving beyond just planning motion to having an immediate reaction capability based on real-time hazard detection while still accounting for those nonlinear liquid dynamics.
Dev: This work on "Model-Predictive Spill-Free Emergency Stopping for Liquid-Manipulating Robots" sets a good standard for how we approach safety in high-speed manipulation tasks.
Taro: It really shows the power of combining established modeling techniques with real-time control to tackle those tricky nonlinear dynamics.
Rosa: Next up, we're going to look at how other papers are tackling the challenges of perception and navigation in complex, real-world settings like these autonomous delivery systems.
Aalto University
cs.RO
Submitted: 2026-04-17
Updated: 2026-11-11
Journal ref: IEEE Robotics and Automation Letters, vol. 11, no. 11, pp. 12751-12758, Nov. 2026
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 70/100
The gist: Manipulating open liquid containers is challenging because liquids are highly sensitive to vessel accelerations and jerks This letter presents an emergency stop system for robots manipulating liquids…
Key concepts
- Optimal Control Problem
- This is a mathematical framework used to find the best sequence of actions (like accelerations) over time. In this paper, it's used to determine the fastest way for a robot carrying a liquid to come to rest without spilling, minimizing velocity while respecting physical limits.
- Model Predictive Control (MPC)
- MPC is a control strategy that uses a mathematical model of the system (like how the liquid sloshes) to predict future behavior over a short time horizon. It repeatedly solves an optimization problem and applies only the first calculated action, allowing for continuous, real-time adaptation to changing conditions during stopping.
- Spherical Pendulum Model
- This is a simplified mechanical model used to simulate how liquid inside a container sloshes or moves when the container tilts. It uses parameters like tilt angles and rod length to predict the liquid's motion, which is crucial for ensuring the robot stops without spilling.
- No-Spill Constraints
- These are mathematical rules that must be followed during motion planning to prevent liquid from overflowing or spilling out of the container. The optimization process explicitly includes these constraints to guarantee that any calculated stopping trajectory keeps the liquid surface within safe limits.
Terminology
Summary
Manipulating open liquid containers is challenging because liquids are highly sensitive to vessel accelerations and jerks This letter presents an emergency stop system for robots manipulating liquids in open containers We formulate emergency stopping as an optimal control problem and solve it in a model predictive control framework to generate time-optimal, spill-free stopping trajectories The proposed method works as a plug-and-play safety layer on top of existing slosh-free motion planning methods, enabling immediate reaction to detected hazards while accounting for nonlinear liquid dynamics We demonstrate, through simulation and on a 7-DoF Franka Emika Panda robot, that the proposed approach achieves fast emergency stopping without spilling
Problem Formulation and Approach
We address the problem of bringing a robotic manipulator arm carrying a liquid container to rest as quickly as possible, without spilling the liquid, after an emergency stop is triggered This means minimizing the cumulative sum of the Cartesian velocities while obeying no-spill constraints for the allowed liquid-surface inclination The proposed method uses a standard model predictive control approach, in which we repeatedly solve optimal control problems with a receding time horizon We use a well-established spherical pendulum model of liquid slosh for its computational efficiency and reasonable modeling accuracy The problem is formulated as a Quadratic Program (QP) to minimize the cumulative sum of the end-effector’s (container’s) Cartesian velocities, while respecting the robot’s kinematic limits and no-spill constraints
Modeling Liquid Dynamics
Mechanical models, such as the spherical pendulum model or mass-spring model, are well-established methods for analyzing and predicting liquid slosh The spherical pendulum model's state is parametrized by two tilt angles (θ, ϕ) about the horizontal axes in the world frame and the pendulum rod length (l) We maintain bilinear coupling terms between z¨p and θ, ϕ so that we end up with bilinear small-angle dynamics To use the pendulum model, the rod length l must be estimated based on the container’s shape and size using natural frequencies of slosh
Methodology Implementation
The state space of our model comprises the pendulum angles and angular velocities, the end-effector’s 6-DoF Cartesian velocities, and the end-effector’s 3-DoF Cartesian orientation, totaling 13 state variables We have six control variables, namely the container’s Cartesian accelerations The optimization is run in a loop in receding horizon fashion with 20 Hz frequency, while a high-frequency real-time control loop reads the solution from shared memory and always communicates the first control action to the robot Until the emergency stop is initiated, we assume that the liquid surface is perpendicular to the container’s vertical axis After the emergency stop is initiated, we switch to integrating the pendulum state using the realized Cartesian accelerations
Experimental Validation and Results
The real-world experiments aimed to validate the proposed method with real hardware and liquid We compared the proposed method with a baseline that does not include slosh modeling The results show that the proposed method kept the maximum slosh essentially within the given limit of 5 degrees, only violating it by practically negligible 0.3 degrees On the other hand, the baseline method significantly exceeded this limit and produced maximum slosh of 22. p. 5
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Enhanced emergency stopping capability for liquid-manipulating robots by formulating emergency stopping as an optimal control problem solved in a model predictive control framework to generate time-optimal, spillfree stopping trajectories. This allows immediate reaction to detected hazards while accounting for nonlinear liquid dynamics, enabling fast emergency stopping without spilling.
-
Integration of a plug-and-play safety layer on top of existing slosh-free motion planning methods enables immediate reaction to detected hazards while accounting for nonlinear liquid dynamics. This feature allows the system to transition seamlessly from planned motion to an emergency stop trajectory based on real-time hazard detection without requiring a complete overhaul of the existing planning infrastructure.
-
The robot can perform rapid stopping in a time-optimal manner while ensuring that no liquid is spilled by minimizing the cumulative sum of Cartesian velocities while obeying no-spill constraints for the allowed liquid-surface inclination. This provides a mathematically optimized path to rest under emergency conditions, balancing speed against spill prevention.
-
The system utilizes a standard model predictive control approach with a receding time horizon and linearizes problematic bilinear terms around a nominal trajectory for each time step in the optimization horizon. This enables real-time tractability without sacrificing modeling power for the complex nonlinear liquid dynamics, allowing continuous adaptation to changing states during stopping.
-
The AI system can generate joint space trajectories by solving another small-scale quadratic program using Resolved Acceleration Control with slack variables to convert task space control actions into joint space commands while respecting joint-wise kinematic limits. This ensures that the planned motion is physically executable by the robot manipulator, even under emergency conditions.
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation