Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees

arXiv:2604.12274 · cs.RO · Submitted 2026-04-14 · 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: "Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees".

Title and authors: Rosa: So we're looking at a paper called "Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees." It sounds like they are trying to figure out how to make these kind of robots walk stably, even when they have those knees.

Dev: Yeah, that's right. The core idea is using a linearized model derived from the nonlinear equations of motion to do future state calculations without having to do all that heavy numerical integration. It seems like they took this technique and applied it specifically to a planar biped robot with knees, which is a step up from just the almost linear ones.

Taro: I'm curious about what "instantaneous walkability determination" actually means in practice for the robot. Is it something that lets us predict if it can keep walking right now?

Rosa: Exactly. They claim they can do this through iterative calculation using this linearized model, which means you just feed in some inputs and see what happens next without running a full simulation every single time. It's about speed and precision in figuring out if the robot will fail immediately or if it can handle the next step.

Dev: From a control standpoint, that sounds like they're simplifying the dynamics significantly to make it tractable for real-time use. They start by deriving the equations of motion and constraint conditions for a planar six degree of freedom biped robot, and then they focus on reducing that down to a three degree of freedom model.

Taro: What about those constraints? You mentioned the stance foot contact not slipping—how does that translate into the math? Is it just setting some velocities to zero?

Rosa: Right. The velocity constraint conditions state that the horizontal position and vertical position of the stance foot must have zero velocity, meaning no slipping at that point. That's a key part of setting up those constraints for stability on the ground, which is described in their derivation for this paper > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: And then they get to the linearization step where they approximate the gravity term as linear around a specific expansion point for the thigh frame angle. That's where they create that reduced linearized model, which looks like M bar double q bar plus G bar q bar plus g bar beta equals S bar u bar > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Title and authors: Taro: That sounds like a lot of assumptions to make when you're trying to model real-world dynamics. How do they justify that approximation, especially since it’s around the thigh frame angle?

Rosa: They found that for these almost linear bipeds, the equations of motion have a constant inertia matrix and no nonlinear velocity terms, which makes approximating the gravity term as linear around that expansion point pretty easy > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: And then they use that linearized model to find the control input u bar using a specific formula involving S bar T M bar inverse Sbar inverse v plus Sbar T Mbar Gbar q bar plus gbar beta > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: So, they're essentially creating a shortcut to predict the next state, which is cool, but what about the actual gait generation part? How do they actually generate that stable walking gait using this model?

Rosa: They realize that the controlled linearized reduced model can be put into state-space form as x dot equals A x plus b1 plus b2 v two plus b3 v three > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: The state at a specific time set in the i th step can then be found by using e A T set x plus i plus eta one plus eta two i plus eta three > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: What's that term eta two? It sounds like it's some kind of correction or integral based on the velocity before impact. That seems important for capturing the impulse effects, right?

Rosa: That term eta two is defined as a definite integral calculation involving v two, which means it’s a function vector containing the angular velocity immediately before the i th impact > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Title and authors: Dev: So, when they analyze the gait descriptors—like step period and angular velocity immediately before impact—they compare them against their nonlinear model, and they found that by setting theta star two to minus zero point two five radians, the step period values are nearly identical > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: But there's a trade-off mentioned; the CLRed model shows slightly larger angular velocity values than the nonlinear model, while the nonlinear one shows significantly larger ones > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: They also tested how gait descriptors change based on beta of the nonlinear model by setting kappa to seven different values, and they found that the case with kappa equal to negative zero point five achieves a very high level of approximation accuracy for both steady step period and walking speed > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: Let's pivot to those descending small steps, because that’s a practical test. They found that when descending a small step, just adjusting the target settling time wasn't enough to generate adaptive walking motion > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: So what did they actually do to fix that issue? Did they change the whole approach, or just tweak one parameter?

Rosa: They proposed modifying T set to be shorter, specifically for the ten th step only, and when T set is set to three values—zero point five five seconds, zero point five seconds, and zero point four five seconds—the robot successfully descends the step and gets back to steady walking > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2> <ref:2604.12274#pg1,Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear>.

Dev: That sounds like a specific tuning solution rather than a general fix for all descent scenarios, which is what I'd expect when you're dealing with real constraints. It highlights how sensitive these gait generators are to those initial parameters > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: So, the big implication here is that by choosing the right expansion point for linearizing that gravity term, they can get a highly accurate CLRed model regardless of what beta is doing > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Title and authors: Rosa: That accuracy allows for instantaneous walkability determination under various conditions, which is the main selling point for this work > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: This method offers an advantage unique to almost linear bipedal robots, allowing them to predict their future states instantly without needing slow numerical integration > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: It's interesting that they noted this robot model doesn't exhibit the natural swinging motion or knee flexion extension movements of the swing leg you see in passive dynamic walkers > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: So, to wrap up, this paper on "Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees" shows that by carefully choosing the expansion point for linearization of the gravity term, you can get a highly accurate CLRed model > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: This capability lets us determine walkability instantaneously, which is a big deal because it means we don't have to wait for long simulations to know if a robot can handle the next move > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Taro: I just want to reiterate that this method relies on an approximate HZD to derive the linearized model, allowing for those fast calculations of future states without numerical integration > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: It's a method that extends the technique to a planar biped robot model with knees, and it demonstrates instantaneous walkability determination through iterative calculation > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: We’ll keep an eye on how this method performs when we push the limits of the system under various conditions > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: I'm looking forward to seeing how they handle real-world failure modes when the world doesn't behave exactly as they modeled > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

The paper's summary: Rosa: So, this paper is basically about taking almost linear biped robots that have knees and making them walk stably in a really smart way—specifically, figuring out if they can walk right now without needing to run a huge simulation for every single step.

Dev: Right, the big idea there is using this linearized model derived from the nonlinear equations of motion to calculate future states instantly instead of doing heavy numerical integration > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: It’s about taking the complexity out of predicting those next steps by using a simplified, linearized model around an expansion point for the thigh frame angle > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: And they use that reduced model to iteratively calculate if the robot can maintain stability by applying control inputs and seeing how the state-space realization behaves during discretization > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: They found that when you compare their simplified, linearized model to the actual nonlinear model, they can get step period values that are nearly identical if you set a specific parameter called theta star two to negative zero point two five radians > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: But there’s a catch, because the linearized model shows slightly larger angular velocity values compared to the nonlinear one, so you have to be careful about that trade-off > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: They also explored how changing the stance-knee angle, which they call beta, affects things by testing seven different values for kappa, and they found that setting kappa to negative zero point five gives them a very high level of accuracy for steady step period and walking speed > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: This means the method works well across different configurations of the nonlinear robot, as long as you pick the right expansion point for that gravity term linearization > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: And they didn't just stop there; they tested descending small steps, and they discovered that you can’t just tweak the settling time anymore; you have to shorten the target settling time specifically for the tenth step to get it to work > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: That shows that even with this high-precision model, real world conditions like descending uneven terrain still require specific adjustments to the parameters, which is a good safety case point > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: The main implication here is that this approach gives almost linear bipeds a way to determine walkability instantly under various conditions without waiting for lengthy numerical integration, which is huge for real-time control systems > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: It’s a neat trick, but you have to remember that this specific paper models a robot that doesn't show the natural swinging motion or knee flexion movements you see in passive dynamic walkers > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: So, the core value is getting this highly accurate CLRed model so we can predict how a robot will move next, even when it hits some tricky situations > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: Next up, we’re going to talk about how these kinds of fast predictions fit into larger systems that have to handle uncertainty, like the ones I work on with reinforcement learning and control barrier functions > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

The paper's improvements: Tom: So, we've seen how they use linearization to get fast predictions, and now we're looking at what they suggest to actually improve the system’s performance when things get messy.

Rosa: They pointed out that while the core method is solid for almost linear bipeds, you still need specific tuning for real-world scenarios like descending a small step > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: Exactly, they showed that just adjusting the target settling time wasn't enough to get adaptive walking motion when those steps got smaller, so the improvement is to modify the target set time > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: They propose a change where you shorten that target set time specifically for the tenth step only, and they tested three different values—zero point five five seconds, zero point five seconds, and zero point four five seconds > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: That shows that the robot’s gait generator is really sensitive to those initial parameters, so the safety case involves finding those specific values for different terrains > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: The big implication here is that it moves from just a theoretical model to something you can actually tune to handle unexpected terrain changes, which is what field trials need > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: But we have to remember the limitation they flagged—this model doesn't capture the natural swinging motion or knee flexion movements you see in passive dynamic walkers, so it’s not a full dynamics solver > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: True, so it’s more about generating the stable walking pattern rather than modeling every tiny muscle twitch of the swing leg > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: This is important because if we use this for real-time control, we need to ensure that when the robot hits something unexpected, it doesn't just stop—it has to recover gracefully > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: So, while the model is accurate for steady walking, its improvement lies in making the descent phase robust enough to handle those small step adjustments reliably > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: It’s about building a control strategy that can adapt to small changes in the environment, which is a necessary step before we try applying this to more complex systems like those we look at in reinforcement learning > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Conclusion: Rosa: So, to wrap things up, we’re looking at the paper "Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees" > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: Basically, this work shows that by carefully choosing the expansion point when you linearize the gravity term, you can get a highly accurate model even when the robot's stance-knee angle, beta, is changing > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: That accuracy means we can determine walkability instantly under various conditions without waiting for a slow numerical integration to finish > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: It’s a big deal because it gives us a tool that lets us check safety right now, which is what I’m always looking for in real-time loops > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: I just want to emphasize that this method is unique to almost linear bipeds because it doesn't show those natural swinging motions or knee flexion you see in passive dynamic walkers > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Rosa: Yeah, it’s a specific robot class, which is important context for how fast this method can be applied to other systems > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Dev: It’s a solid result because they showed that by modifying the target set time for the tenth step, you can handle descending small steps successfully > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: So, as we look at more complex autonomy research, this is a good starting point because it proves that fast prediction is possible even when you have nonlinear dynamics on your side > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Rosa: Exactly. It sets a baseline for how much faster we can get at gait synthesis using these linearized approximations > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Dev: Alright, that's the wrap on this paper; it’s interesting work showing how mathematical approximation can speed up control decisions without sacrificing too much accuracy on a specific robot type > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg2>.

Taro: Next time, I want to see if we can push this kind of predictive modeling into those more complex, stochastic environments we’re testing in the autonomous navigation papers > Asymptotically Stable Gait Generation and Instantaneous Walkability Determination for Planar Almost Linear Biped with Knees <ref:2604.12274#pg1>.

Japan Advanced Institute of Science and Technology

cs.RO

Submitted: 2026-04-14

Updated: 2026-11-15

Comments: Accepted for presentation at the IEEE International Conference on Robotics and Automation (ICRA), 2026. The previous version corrected a typographical error in one equation. This version corrects the author names in a reference. The scientific content remains unchanged

Journal ref: 2026 IEEE International Conference on Robotics and Automation (ICRA), pp. 598-604, 2026

DOI: 10.1109/ICRA57385.2026.11697158

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

Importance score: 76/100

The gist: Asymptotically stable gait generation and instantaneous walkability determination for planar almost linear biped with knees is a method that allows for high-precision future state calculation without…

Key concepts

Almost Linear Biped
This refers to a type of bipedal robot where the equations describing its movement are nearly linear. This property simplifies complex dynamics because the inertia matrix is constant, and nonlinear gravity terms can be easily approximated as linear, making mathematical modeling much more tractable.
Linearized Model (CLRed)
The original nonlinear equations of motion are simplified into a reduced, linearized model. This approximation is done by treating the gravity term linearly around a specific point (the expansion point). This allows the system to be represented in a standard state-space form, enabling rapid calculations of future states without needing slow numerical integration.
Instantaneous Walkability Determination
This is the ability to immediately assess whether a robot can successfully walk under current conditions. The method uses iterative calculations on the linearized model and its discretized form to quickly predict gait stability and performance, which is much faster than simulating every step numerically.
Gait Descriptors Comparison
The study compares metrics like step period and angular velocity immediately before impact between the robot's complex nonlinear model and the simplified CLRed model. By testing different parameters, researchers found that setting a specific parameter ($eta$) yields very accurate predictions for these gait characteristics.

Terminology

Summary

Asymptotically stable gait generation and instantaneous walkability determination for planar almost linear biped with knees is a method that allows for high-precision future state calculation without numerical integration by utilizing a linearized model derived from the nonlinear equations of motion This paper extends this technique to a planar biped robot model with knees, demonstrating instantaneous walkability determination through iterative calculation. p. 0

Model Derivation and Constraints

The study begins by deriving the equations of motion, constraint conditions, and inelastic collisions for a planar 6-DOF biped robot. The robot's equation of motion is given by Mq¨ + Cq˙ + g = Su + J T c λc. A key feature of these almost linear bipedal robots is that their equations of motion have a constant inertia matrix, no nonlinear velocity term, and nonlinear terms in the gravity term that can be easily approximated as linear. p. 0 The velocity constraint conditions state x˙ = 0 and z˙ = 0 for the stance foot contact without slipping. p. 1

Linearization and Reduction

The equations of motion are reduced to a 3-DOF model by considering the velocity constraint condition in Eq. (4) which yields q˙ = 0. p. 2 A linearized model is derived by approximating the gravity term as linear around the expansion point for the thigh frame angle. p. 0 The reduced linearized model is obtained as M¯ ¨¯q + G¯ q¯ + g¯β = S¯ū. p. 3 This reduced linearized model allows for the determination of the control input ū using the formula ū = S¯ TM¯−1 Sbar−1 v + Sbar TM¯−1 Gbar q̄ + gbarβ. p. 3

Discretization and State Realization

The controlled linearized reduced (CLRed) model is realized in state-space form as x˙ = Ax + b1 + b2v2 + b3v3. The state at t = Tset in the i-th step can be obtained as x(Tset) = eATset x + i + η1 + η2i + η3. p. 4 The term η2 is defined as a definite integral calculation involving v2, meaning it is a function vector containing the angular velocity immediately before the i-th impact. p. 4

Gait Analysis and Accuracy

The comparison of gait descriptors between the nonlinear and CLRed models shows that the CLRed model with θ∗2 = -0.25 [rad] shows step period values nearly identical to those of the nonlinear model. p. 5 The former shows slightly larger angular velocity values than the nonlinear model, while the latter shows significantly larger values. p. 5 Analyzing gait descriptors with respect to β of the nonlinear model and CLRed models by setting κ to seven different values reveals that the case with κ = −0. p. 55 achieves a very high level of approximation accuracy for the values of steady step period and walking speed. p. 5

Descending Small Steps

When descending a small step, it was revealed that continued walking becomes impossible, and that adjusting only the target settling time could not generate sufficiently adaptive walking motion. The authors propose modifying Tset to be shorter T′set only for the 10-th step and analyze its effect below. p. 6 This modification shows that when T′set is set to three values, 0. p. 655, 0.50 and 0.45 [s], the robot successfully descends the step and returns to steady walking. p. 6

Conclusion

The paper demonstrates that by appropriately setting the expansion point for linearization of the gravity term, a highly accurate CLRed model can be obtained regardless of the value of β. This advantage unique to AL3 robots will enable instantaneous walkability determination under various conditions. p. 6 The authors anticipate that this method will enable instantaneous walkability determination under various conditions. p. 6 The biped robot model discussed in this paper does not exhibit the natural swinging motion or knee flexion/extension movements of the swing leg seen in passive dynamic walkers. p. 6

REFERENCES

[1] T. McGeer, “Passive dynamic walking,” Int. p. 6 J. of Robotics Research, Vol. p. 6 9, Iss. 2, pp. 62–82, 1990 <ref:2604. p. 0 p. 012274#pg2>

[3] J. W. Grizzle, G. p. 6 Abba and F. p. 6 Plestan., “Asymptotically stable walking for biped robots: Analysis via systems with impulse effects,” IEEE Trans. p. 6 on Automatic Control, Vol. p. 6 46, Iss. 1, pp. 51–64, 2001 <ref:2604. p. 012274#pg4>

[5] J. Kiefer and B. p. 0 Ramesh., “Walking viability and gait synthesis for a novel class of dynamically-simple bipeds,” Informatica, Vol. p. 6 17, Iss. 2, pp. 145–155, 1993

[6] M. W. Spong, R. Lozano and R. p. 6 Mahony., “An almost linear biped,” Proc. p. 6 of the 39th IEEE Conf. p. 6 on Decision and Control, pp. p. 6 4803–4808, 2000

[7] S. K. Agrawal and A. p. 0 Fattah., “Motion control of a novel planar biped with nearly linear dynamics,” IEEE/ASME Trans. p. 6 on Mechatronics, Vol. p. 6 p. 6 11, Iss. 2, pp. 162–168, 2006

[8] V. Sangwan and S. p. 6 K. Agrawal., “Differentially flat design of bipeds ensuring limit cycles,” IEEE/ASME Trans. p. 6 on Mechatronics, Vol. 14, Iss. 6, pp. 647–657, 2009

[9] F. Asano., “A novel generation method for underactuated bipedal gait with landing position control of swing leg based on property of zero dynamics,” Proc. p. 6 of the IEEE Int. p. 6 Conf. on Advanced Intelligent Mechatronics, pp. p. 6 1184-1189, 2015

[10] J. Li and Q. Nguyen., “Dynamic walking of bipedal robots on uneven stepping stones via adaptive-frequency MPC,” IEEE Control Systems Letters, Vol. p. 6 7, pp. 1279–1284, 2023

[11] F. Jenelten, R. p. 6 Grandia, F. p. 6 Farshidian and M. p. 6 Hutter., “TAMOLS: Terrain-aware motion optimization for legged systems,” IEEE Trans. p. 6 on Robotics, Vol. p. 6 38, Iss. 6, pp. 3395–3413, 2022

[1] T McGeer, “Passive dynamic walking,” Int J of Robotics Research, Vol 9 Iss 2, pp. 62–82, 1990 <ref:2604.12274#pg2>

[5] J Kiefer and B Ramesh., “Walking viability and gait synthesis for a novel class of dynamically-simple bipeds,” Informatica, Vol 17 Iss 2, pp. p. 6 145–155, 1993 <ref:2604. p. 612274#pg3>

[6] M W Spong, R Lozano and R Mahony., “An almost linear biped,” Proc of the 39th IEEE Conf on Decision and Control, pp. p. 6 4803–4808, 2000 <ref:2604. p. 612274#pg2>

[7] S K Agrawal and A Fattah., “Motion control of a novel planar biped with nearly linear dynamics,” IEEE/ASME Trans on Mechatronics, Vol 11 Iss 2, pp. p. 6 162–168, 2006 <ref:2604. p. 612274#pg4>

[8] V Sangwan and S K Agrawal., “Differentially flat design of bipeds ensuring limit cycles,” IEEE/ASME Trans on Mechatronics, Vol 14 Iss 6, pp. p. 6 647–657, 2009 <ref:2604. p. 612274#pg4>

[9] F Asano., “A novel generation method for underactuated bipedal gait with landing position control of swing leg based on property of zero dynamics,” Proc of the IEEE Int Conf on Advanced Intelligent Mechatronics, pp. p. 6 1184-1189, 2015 <ref:2604. p. 612274#pg4>

[10] J Li and Q Nguyen., “Dynamic walking of bipedal robots on uneven stepping stones via adaptive-frequency MPC,” IEEE Control Systems Letters, Vol 7, pp. p. 6 1279–1284, 2023 <ref:2604. p. 612274#pg4>

[11] F Jenelten, R Grandia, F Farshidian and M Hutter., “TAMOLS: Terrain-aware motion optimization for legged systems,” IEEE Trans on Robotics, Vol 38 Iss 6, pp. p. 6 3395–3413, 2022 <ref:2604. p. 6

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. Improved AI systems can perform instantaneous walkability determination by utilizing an approximate HZD to derive a linearized model, allowing calculations for future states without numerical integration.

  2. The improved system can achieve high-precision future state calculations by reducing the equations of motion to a 3-DOF model and deriving a linearized model by approximating the gravity term as linear around the expansion point for the thigh frame angle.

  3. The system can perform iterative calculation for instantaneous walkability determination through applying control inputs to the obtained model, performing state-space realization, and then discretizing it.

  4. The system can generate asymptotically stable gaits on a horizontal plane by designing and applying control inputs that achieve y¨ = v using v = y¨d(t) as the acceleration command vector.

  5. The improved system can accurately predict angular positions, velocities, control inputs, ground reaction forces, and swing foot position during gait generation through numerical simulations.

  6. The system can analyze gait descriptors like step period and angular velocity immediately before impact using the CLRed model to compare against nonlinear models for faster analysis than traditional numerical integration.

  7. The system can investigate changes in gait characteristics corresponding to the stance-knee angle, beta, by utilizing the CLRed model with different expansion points for linearization of the gravity term.

  8. The system can assess the feasibility of descending a small step by analyzing gait descriptors during generated gaits involving descending 2.0 cm steps and adjusting target settling time accordingly.

Related papers