Organizations: Graduate School of Information Science, University of Hyogo, Hyogo, Japan · Takenaka Research & Development Institute, Takenaka Corporation, Chiba, Japan
Exploring lunar lava tubes requires robots to traverse without human intervention. Because pre-trained policies cannot fully cover all possible terrain conditions, our goal is to enable adaptive policy switching, where the robot selects an appropriate terrain-specialized model based on its current terrain features. This study investigates whether terrain types can be estimated effectively using posture-related observations collected during navigation. We fine-tuned a pre-trained policy using Proximal Policy Optimization (PPO), and then collected the robot's 3D orientation data as it moved across flat and rough terrain in a simulated lava-tube environment. Our analysis revealed that the standard deviation of the robot's pitch data shows a clear difference between these two terrain types. Using Gaussian mixture models (GMM), we evaluated terrain classification across various window sizes. An accuracy of more than 98% was achieved when using a 70-step window. The result suggests that short-term orientation data are sufficient for reliable terrain estimation, providing a foundation for adaptive policy switching.
Figures & tables
Table 1: Target-reaching success rates and average target-reaching times with the combination of models and terrains [ 10 ] . A general model is trained in both flat and rough areas.
Figure 1: Training process for terrain-specialized models
Figure 2: Simulation environment
Figure 3: Robot with defined rotation axes
Figure 4: Definitions of the distances
Hyperparameter
Initial Flat
General
Total Steps
2.5M
2.5M
Batch Size
64
64
Epochs
10
10
LR
3×10−4
3×10−4
ϵ
0.2
0.2
γ
0.99
0.99
Table 2: Hyperparameters
Figure 5: sinθz (roll) in the two areas
Figure 6: sinθx (pitch) in the two areas
Figure 7: Distribution of sinθx std. (100-step window)
Window size
Mean (Std.)
Accuracy
10
0.02 (0.01) / 0.08 (0.04)
61.13 %
20
0.03 (0.01) / 0.08 (0.03)
70.73 %
40
0.05 (0.01) / 0.11 (0.07)
85.87 %
70
0.05 (0.01) / 0.11 (0.03)
98.79 %
Table 3: GMM clustering performance by window size. The middle column displays the estimated mean and standard deviation for the flat / rough clusters, respectively.