Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints

arXiv:2603.04757 · cs.RO · Submitted 2026-03-05 · Read on arXiv

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: "Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints".

Title and authors: Rosa: We started by talking about the paper "Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints." The title itself really tells you the core problem they are tackling: balancing joint load and mobility in modular robots, especially when those joints can be easily detached.

Dev: And the authors, Kennosuke Chihara, Takuya Kiyokawa, and Kensuke Harada. They are clearly experts in both the control side of things and the physical robot design itself.

Taro: I’m always curious about what they were focusing on specifically with those detachable joints. Were they looking at how that physical locking structure impacts the optimization?

Rosa: Yes, they focused on how those detachable joints, like the Twister or Pivot mentioned in their hardware description, interact with the load minimization objective. They used a simple detachment mechanism with a physical locking structure using a coil spring and protrusion to achieve both easy detachment and high rigidity.

Dev: That physical aspect is crucial because if you're optimizing for load, you have to know exactly how that mechanical part behaves under stress when it’s engaged versus when it’s detached.

Taro: So, were they running simulations where they could explicitly model the torque spikes that happen right at the moment of detachment or engagement? That would be a big test for their framework.

Rosa: They did perform simulations to verify this, and the core idea is that motion generation considering joint load is essential because even with high-rigidity joints, there's still a risk of damage in high-load tasks like slope climbing.

Dev: That leads right into the optimization problem they set up. They are searching for gait parameters using a multi-objective optimization algorithm to achieve both hardware protection and environmental adaptation simultaneously.

Taro: How did they structure that search? Were they using a brute force method, or something more sophisticated when you're dealing with these complex constraints?

Rosa: They used the NSGAIII algorithm for this search, which is a multi-objective optimization algorithm they employed to generate the gait parameters.

Dev: NSGAIII is designed for finding Pareto optimal solutions in problems with many objectives, and here they have three objectives: locomotion speed, stability, and joint load. That’s a tough trade-off to manage without getting stuck in a local optimum.

Taro: So they are trying to find the best possible compromise among those three competing goals simultaneously across different terrains like flat ground, slopes, and steps.

Rosa: That's right. The goal isn't just to maximize speed; it’s to derive a diverse set of Pareto optimal solutions that explicitly balance mobility and structural integrity as they say in their abstract > Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints Kennosuke Chihara, Takuya Kiyokawa, and Kensuke Harada twelve Abstract—While modular robots offer versatility, excessive joint torque during locomotion poses a significant risk of mechanical failure, especially for detachable joints. To address this, we propose an optimization framework using the NSGAIII algorithm. Unlike conventional approaches that prioritize mobility alone, our method derives Pareto optimal solutions to minimize joint load while maintaining necessary locomotion speed and stability. Simulations and physical experiments demonstrate that our approach successfully generates gait motions for diverse environments, such as slopes and steps, ensuring structural integrity without compromising overall mobility <ref:2603.04757#pg2>.

Dev: So the setup is clear: they have a defined problem with three conflicting goals, and they are using a powerful algorithm to find solutions that sit on the Pareto front where you can’t improve one objective without hurting another.

Taro: It sounds like the real test is seeing if those generated gaits actually work reliably when the world throws things at them, rather than just looking good in a simulation.

The paper's summary: Rosa: Now let's talk about what this paper actually summarizes. Essentially, they are showing that while modular robots are versatile, they need a better way to manage joint stress during movement because excessive torque poses a real risk of mechanical failure for those detachable joints.

Dev: They propose an optimization framework that moves away from prioritizing mobility alone and instead derives Pareto optimal solutions. This means the system is designed to minimize joint load while still maintaining necessary locomotion speed and stability requirements.

Taro: So, the summary is essentially: we need a system that protects the hardware by actively minimizing joint load while ensuring the robot can still move effectively in its intended environment.

Rosa: That’s a good way to put it. The paper summarizes their approach to generating gait motions for diverse environments like slopes and steps, proving that their method ensures structural integrity without compromising overall mobility.

Dev: They demonstrate this through simulations and physical experiments on both four-legged and six-legged robot models across various terrains. This validates that the proposed method can generate functional gaits for different morphologies.

Taro: So, they aren't just claiming it works in one perfect scenario; they are testing its adaptive capabilities by environment and gait, which is what we need when a robot is actually operating outside the lab.

Rosa: That’s the point. They are showing that their framework allows for adaptive capabilities by environment and gait, enabling the robot to dynamically reconfigure its morphology, for instance, deploying four legs on slopes, or switching gaits like using a tripod gait for steps based on what works in that specific terrain.

Dev: And they also provide a statistical analysis of factors affecting joint load. They identified swing height H as having the largest partial regression coefficient of zero point four zero one with high statistical significance when operating on flat terrain, which points to it being a key factor in load management there.

The paper's improvements: Taro: So if we look at what the authors suggest for improvements, they’re suggesting that we need to be more granular about the design changes based on the environment and gait.

Rosa: Exactly. They suggest that by analyzing which gaits succeed or fail in specific environments, you can derive insights into how to adapt your morphology and switch gaits based on terrain demands.

Dev: And they also highlight that the system can identify swing height H as a dominant factor, which means you should target reducing that swing height to protect modular interfaces under high load on flat terrain where it shows a statistically significant positive correlation with joint load.

Taro: That’s a concrete piece of actionable advice for designers: focus your efforts on suppressing that swing height to keep the joints safe in those specific situations.

Rosa: And they also admit the sim-to-real gap, which is caused by body sinking due to elastic deformation of three dee-printed parts and joint backlash <ref:2603.04757#pg2>. They suggest future work should focus on improving hardware rigidity or modeling friction characteristics more accurately for that real-world performance issue.

Dev: So the improvement path isn't just about tweaking the algorithm; it's also about addressing those physical realities of manufacturing and making sure the robot can actually handle them physically.

Taro: I agree, because if you don't fix the hardware rigidity or friction modeling, all that optimization work is still constrained by those physical imperfections.

Conclusion: Rosa: So to wrap up on "Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints," the paper presents a framework that successfully balances mobility and structural integrity through multi-objective optimization.

Dev: The main implication is that you can now use this framework to generate gait motions for diverse environments, ensuring structural integrity without compromising overall mobility.

Taro: It gives us a tool to generate gaits where the performance is optimized against both load and speed metrics simultaneously, which is something we needed when we were testing real-world failure modes.

Rosa: And the next steps point towards using this framework to adapt robot morphology and switch gaits based on terrain demands, while also tackling those simulation-to-real gaps with better hardware or more accurate friction modeling.

Dev: It’s a solid method for ensuring mechanical reliability in systems with modular components where you need high performance under stress.

Taro: Yeah, it gives us a way to think about how the robot should gracefully recover when things go wrong during locomotion.

Rosa: That's all we have for this paper today on "Gait Generation Balancing Joint Load and Mobility for Legged Modular Robots with Easily Detachable Joints."

Kennosuke Chihara, Takuya Kiyokawa, Kensuke Harada

Department of Systems Innovation, Graduate School of Engineering Science, The University of Osaka · Industrial Cyber-physical Systems Research Center, The National Institute of Advanced Industrial Science and Technology

cs.RO

Submitted: 2026-03-05

Updated: 2026-11-16

Comments: 8 pages, 9 figures

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 78/100

The gist: Modular reconfigurable robots require optimization frameworks that balance locomotion speed, stability…

Key concepts

Objective Function J(O)
This is the mathematical goal the system tries to minimize. It combines three factors: locomotion speed, stability, and joint load. The system seeks a set of robot parameters (like leg speeds and heights) that results in the best possible balance among these competing goals.
Pareto-optimal Solutions
These are solutions where you cannot improve one objective without making another objective worse. Instead of choosing just the fastest or lightest gait, this method finds a set of compromises—solutions that explicitly trade off mobility against structural integrity and joint stress.
Joint Load Metric (f_load)
This measures how much force is exerted on the robot's joints during movement. It is calculated by averaging the total reaction force across all joints during a gait cycle and normalizing it by the robot's total weight, helping designers understand mechanical stress.
Swing Height (H) Correlation
Statistical analysis showed that swing height has a strong positive link to joint load on flat ground. This means taller leg swings tend to increase the stress on the joints. Designers can use this finding to reduce swing height if minimizing joint load is a primary concern.

Terminology

Summary

Modular reconfigurable robots require optimization frameworks that balance locomotion speed, stability, and joint load to ensure structural integrity during movement <ref:2603. p. 504757#pg2> This multi-objective approach derives a diverse set of Pareto-optimal solutions enabling an explicit balance between mobility and structural integrity <ref:2603. p. 004757#pg2>.

Problem Formulation

The optimization problem is formulated to minimize the objective function J(O) which includes locomotion speed, stability, and joint load <ref:2603.04757#pg4>. The decision variable O is defined as a set of parameters that define the leg trajectory shape, speed, and timing <ref:2603. p. 104757#pg4>. This set includes stride lengths Li and swing speeds Vi for each leg, along with the common swing height H and duty factor β <ref:2603. p. 104757#pg4>.

Objective Function Definitions

The three objective functions are defined to capture the desired trade-offs <ref:2603.04757#pg5>.

  1. The speed function f speed is defined by normalizing the locomotion velocity relative to a reference speed <ref:2603. p. 104757#pg6>. It is calculated as ˆvfwd − λ · vˆ2 drift <ref:2603. p. 104757#pg6>.

  2. The stability metric f stability is based on the static stability margin, defined as the time average of the normalized distance between the Center of Mass (CoM) and the boundary of the support polygon. p. 1

  3. The joint load metric f load is defined as the average total joint reaction force during the gait cycle, normalized by robot weight Wrobot. p. 1

Experimental Validation and Results

The framework was validated through simulations and physical experiments using 4-legged and 6-legged robots across various terrains <ref:2603.04757#pg2>. Comparisons showed that the proposed method reduced the maximum joint load by approximately 11. p. 35% compared to the comparative method when evaluating the 6-legged tetrapod gait on flat terrain. p. 2 However, this reduction came at a cost, as locomotion speed decreased by approximately 10. p. 35%.

Analysis of Key Factors

Statistical analysis revealed that on flat terrain, both the duty factor and the swing height had a statistically significant positive correlation with joint load. Specifically, the swing height H exhibited the largest partial regression coefficient (B = 0. p. 4401) with high statistical significance (p = 0. p. 5004 < 0.05). Conversely, all variables related to stride length and swing speed of individual legs did not meet the significance level (p > 0. p. 405).

Sim-to-Real Discrepancies

The reduced locomotion speed observed in physical experiments is primarily attributed to body sinking caused by the elastic deformation of 3D-printed parts and joint backlash. Despite this, successful step traversal confirmed the fundamental effectiveness of the framework, showing that generated Pareto-optimal solutions can ensure both structural integrity and mobility in real-world environments. p. 5

Conclusion

This study presented a multi-objective gait generation framework for modular robots focusing on the trade-off between joint load and locomotion performance. The optimized kinematic parameters clearly demonstrate the method’s ability to ensure structural integrity while maintaining mobility for each morphology. p. 5 Future work will address the Sim-toReal gap through improved hardware rigidity and integrate dynamic stability metrics to explore higher mobility under minimized mechanical stress. p. 5

ACKNOWLEDGEMENT

This work was supported by JSPS KAKENHI Grant Number JP25K21314.

REFERENCES

[1] K. Gilpin and D. p. 5 Rus, “Modular robot systems,” IEEE Robotics & Automation Magazine, vol. p. 5 17, no. 3, pp. 38–55, 2010

[2] M. Yim, W.-M. Shen, B. Salemi, D. Rus, M. Moll, H. Lipson, E. Klavins, and G. p. 5 S. Chirikjian, “Modular self-reconfigurable robot systems [grand challenges of robotics],” IEEE Robotics & Automation Magazine, vol 14 no 1 pp 43–52

[3] S. Murata, E. Yoshida, A. p. 5 Kamimura, H. p. 5 Kurokawa, K. p. 5 Tomita, and S. p. 5 Kokaji, “M-tran: Self-reconfigurable modular robotic system,” IEEE/ASME Trans Mechatronics vol 7 no 4 pp 431–441

[4] A. Zhao, J. Xu, M. Konakovic-Luković, J. p. 5 Hughes, A. Spielberg, D. p. 5 Rus, and W. p. 5 Matusik, “RoboGrammar: graph grammar for terrainoptimized robot design,” ACM Trans Graphics vol 39 no 6 2020

[5] E. Romiti, N. Kashiri, J. p. 5 Malzahn, and N. p. 5 Tsagarakis, “Minimumeffort task-based design optimization of modular reconfigurable robots,” in Proc IEEE Int Conf Robot Autom (ICRA), 2021 pp 9891–9897

[6] M. Mayer and M. p. 5 Althoff, “Holistic optimization of modular robots,” IEEE Trans Automation Science and Engineering vol 23 pp 2703–2716, 2026

[7] A. Sproewitz, A. p. 5 Billard, P. p. 5 Dillenbourg, and A. p. 5 J. Ijspeert, “Roombots-mechanical design of self-reconfiguring modular robots for adaptive furniture,” in Proc IEEE Int Conf Robot Autom (ICRA), 2009 pp 4259–4264

[8] K. Uno, E. Neppel, G. H Diaz, A. Mishra, S. Karimov, A. p. 5 S Jain, A Habib, P Pama, H Gozbasi, S Santra et al. p. 5, “Moonbot: Modular and on-demand reconfigurable robot towards moon base construction,” IEEE Trans Field Robotics 2025

[9] T. Makabe, K. Okada, and M. p. 5 Inaba, “Development of the assembling system for structure transformable humanoid with attach-lockdetachable magnetic coupling,” in Proc IEEE Int Conf Robot Autom (ICRA), 2024 pp 2433–2439

[10] G. Lan, M. De Carlo, F. p. 5 van Diggelen, J. p. 5 M Tomczak, D. p. 5 M Roijers, and A E Eiben, “Learning directed locomotion in modular robots with evolvable morphologies,” Applied Soft Computing vol 111 p 107688 2021

[11] J. Lee, J. Hwangbo, L. p. 5 Wellhausen, V Koltun, and M Hutter, “Learning quadrupedal locomotion over challenging terrain,” Science robotics vol 5 no 47 p eabc5986 2020

[12] S. Choi, G. Ji, J Park, H Kim, J Mun, J H Lee, and J Hwangbo, “Learning quadrupedal locomotion on deformable terrain,” Science Robotics vol 8 no 74 p eade2256 2023

[13] T. Miki, J. Lee, L Wellhausen, and M Hutter, “Learning to walk in confined spaces using 3d representation,” in Proc IEEE Int Conf Robot Autom (ICRA), 2024 pp 8649–8656

[14] Z. Xu, A H Raj, X Xiao, and P Stone, “Dexterous legged locomotion in confined 3d spaces with reinforcement learning,” in Proc IEEE Int Conf Robot Autom (ICRA), 2024 pp 11 474–11 480

[15] R Hoeller, N Rudin, D Sako, and M Hutter, “ANYmal parkour: Learning agile navigation for quadrupedal robots,” Science Robotics vol 9 p eadi7566 2024

[16] H Wang, R Zhou, L Ding, T Liu, Z Zhang, P Xu, H Gao, and Z Deng, “Whole-body constrained learning for legged locomotion via hierarchical optimization,” IEEE Robot Autom Lett. 2025

[17] Y Kim, H Oh, J Lee, J Choi, G Ji, M Jung, D Youm, and J Hwangbo, “Not only rewards but also constraints: Applications on legged robot locomotion,” IEEE Trans Robotics vol 40 pp 2984–3003 2024

[18] S Gangapurwala, M Geisert, R Orsolino, M Fallon, and I Havoutis, “Rloc: Terrain-aware legged locomotion using reinforcement learning and optimal control,” IEEE Trans Robotics vol 38 no 5 pp 2908–2927 2022

[19] A Belmonte-Baeza, J Lee, G Valsecchi, and M Hutter, “Meta reinforcement learning for optimal design of legged robots,” IEEE Robot Autom Lett.

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

  1. Minimizing joint load prevents structural failure during locomotion, ensuring mechanical reliability for modular robots by actively working to minimize this newly introduced load objective as defined in Equation 4. This allows the AI system to generate gait motions that maintain structural integrity across different configurations and diverse terrains, as validated by simulations and experiments.

  2. The system can derive Pareto optimal solutions that explicitly balance mobility and structural integrity rather than prioritizing mobility alone, addressing the limitation of conventional approaches that typically prioritize mobility only. This enables the AI to select gaits based on a trade-off between locomotion speed, stability, and joint load simultaneously.

  3. The framework allows for adaptive capabilities by environment and gait, enabling the robot to dynamically reconfigure its morphology (e.g., deploying four legs on slopes) and switch gaits (e.g., using the tripod gait for steps) based on terrain demands. This capability is informed by analyzing which gaits succeed or fail in specific environments, as summarized in Table III.

  4. The system can identify the most dominant factor affecting joint load, which is swing height H, as shown by the largest partial regression coefficient (B = 0.401). This allows for targeted design modifications to suppress swing height to protect modular interfaces under high load on flat terrain, where it exhibits a statistically significant positive correlation with joint load.

  5. The AI system can predict and mitigate Sim-to-Real gaps by recognizing that reduced locomotion speed in physical experiments is primarily due to body sinking caused by the elastic deformation of 3D-printed parts and joint backlash. This suggests future improvements should focus on enhancing hardware rigidity or modeling friction characteristics more accurately.

Related papers