The effect of limb kinematics on the speed of a legged robot on granular media

The gist The effect of limb kinematics on the speed of a legged robot on granular media was studied by systematically analyzing SandBot's performance as a function of gait parameters to determine optimal locomotion strategies for yielding terrain.

Background and Physics of Interaction

Locomotion performance on granular media is controlled by the relative magnitude of the penetration resistance force and the sum of external forces like weight and inertial forces, which determines whether the granular media solidifies or flows The previous study revealed that as a limb vertically penetrates into the medium, the penetration force scales with z, the depth of the intruder below the surface as Fp(z) = k(φ)z The study focuses on testing SandBot on granular media of different yielding properties set by granular volume fraction at various limb frequencies while keeping other gait parameters fixed <ref:1303.7276#pg3>.

Locomotor Modes and Kinematic Model

The kinematic model describes the limb-ground interaction of SandBot by considering the motion of just a single limb, where penetration force can be written as Fp(θ) = 2Rk cos θ - h / 2R <ref:1303.7276#pg4>. Two distinct locomotor modes are possible depending on whether or not Fp(θ = 0) > m(g + a) <ref:1303.7276#pg5>.

1. Rotary walking – movement with solidification occurs when Fp and m(g + a) intersect, leading to leg penetration stopping at an angle θ = βi when Fp(βi) = m(g+a) <ref:1303.7276#pg5>. This mode involves the c-leg rotating about its center of curvature lifting and propelling the body forward by a step length s = R(sin βf −sin βi) <ref:1303.7276#pg5>.

2. Swimming – movement without solidification occurs when Fp(0) < m(g+a), where the limb constantly slips through fluidized granular material, resulting in a slow speed vx.

Effect of Gait Parameters on Speed

The robot speed depends sensitively on limb kinematics, and the study systematically varied gait parameters including angular extent (θs), angular location (θ0), and temporal duty factor (dc) <ref:1303.7276#pg6>. For fixed ω = 8 rad/s and φ = 0.605, the experimental data showed a single sharp peak in speed near the parameters θs = 1.5, θ0 = −0.5 <ref:1303.7276#pg6>. The optimal gait parameters found were set to be exactly these values plus dc = 0.55 <ref:1303.7276#pg6>.

Anisotropic Penetration Force Law

The model was modified to incorporate an anisotropic penetration force during rotational intrusion, using the law Fp(θ) = 2Rk0 / [cos [b (θ − β0)] + 1] for F > 0 <ref:1303.7276#pg7>. This modification allowed the model to capture the shift in peak location to θ0 = −0.5, which was observed experimentally as an asymmetry to θ0 = 0 <ref:1303.7276#pg7>. The anisotropy in penetration force during rotation is speculated to be a result of changing limb orientation during rotational intrusion <ref:1303.7276#pg9>.

Effect of Duty Cycle and Overlap

Variation of the duty cycle at fixed θs and θ0 also has a substantial influence on speed, with data showing a well-defined peak at dc ≈ 0.5 <ref:1303.7276#pg7>. The model prediction for speed vs dc showed discrepancies in regions I, II, and III due to swimming modes or tripod overlap <ref:1303.7276#pg9>. Tripod overlap at high dc is predicted to halve the displacement from 2s to s <ref:1303.7276#pg9>. The results suggest that angular extent and location of the slow phase must be adjusted to optimize interaction with granular media by minimizing inertial force and limb interference <ref:1303.7276#pg10>.

Conclusion

The study concludes that gait parameters controlling the angular extent and location of the slow phase must be adjusted to optimize interaction with granular media by minimizing inertial force and limb interference, and maximizing the use of solid properties of granular media. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media. Future work should compare these results to investigations of gait optimization on hard ground. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce <ref:1303.7276#pg10>. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through mechanical feedback.

1. For legged robots operating on granular media, AI systems can optimize gait parameters by targeting a speed optimum in gait parameter space by controlling angular extent and angular location of the slow phase to minimize inertial force and limb interference, and maximizing the use of solid properties of granular media.

2. AI-driven control systems can predict locomotion mode (rotary walking vs. swimming) by dynamically adjusting parameters based on the penetration force balance: When Fp(z) and m(g + a) intersect rotary walking occurs. When Fp and m(g + a) do not intersect (dashed red curve and above), the robot swims.

3. AI models can incorporate an anisotropic penetration force law to better predict performance, specifically by modifying the model's force law from Fp(z) = kz to Fp(θ) = 2Rk0 cos [b (θ − β0)] + 1」where β0 = −0.75, leading to a prediction that captures the shift in peak location to θ0 = −0.5 (i.e. asymmetry to θ0 = 0).

4. AI systems can optimize duty cycle control based on granular media characteristics, as the paper shows that "Speed drops off relatively slowly for dc > 0.5, and more quickly to small (swimming) speeds for dc < 0.5."