Robot foundation models may generalize across tasks and embodiments, but their actions still depend on sensors that can fail. When the policy is frozen or inaccessible, can an external monitor diagnose and repair corrupted inputs without access to its internals? We test a structured surrogate that separates periodic gait from sensor-dependent corrections to guide input repair around a frozen locomotion policy. On simulated Go2, offline diagnosis raises walking success from 259/640 to 429/640 episodes; online repair reaches 382/640, with no triggers in 64 held-out fault-free episodes. A matched boosted-tree surrogate predicts clean actions similarly well but yields no detected repair improvement over no repair. Separate fits improve walking on Go1 and ANYmal C, although phase-aware gradients perform better. Evidence remains limited to one policy and a partly screened simulated fault suite; robot foundation model and hardware validation remain open.
Figures & tables
Figure 1: Method overview: Diagnosis and input repair. MBD identifies the faulty sensor from the structured surrogate’s per-sensor contributions, and PRR repairs its input before the frozen policy acts. For a +0.3 rear-left thigh-encoder offset: (a) the sensor explains 49% of the abnormal correction across 16 diagnosis logs; (b) PRR supplies its phase-matched healthy value; and (c) repair removes the sideways drift and raises walking success from 3/16 to 16/16 fresh episodes.
Go2
Walk
Method
R2
Diagnosed
Walk
Survive
Go1
ANYmal C
No repair
–
–
259
373
197
162
MBD + PRR
0.9766
27
429
467
305
265
Online MBD + PRR
–
–
382
419
–
–
GBM–TreeSHAP + PRR
0.9764
5
277
361
266
239
Phase-aware gradient + PRR
–
33
408
455
319
283
Table 1: Results across three embodiments. For Go2, R2 measures action prediction on 16 clean trajectories, and Diagnosed reports top-1 sensor identification across 37 faults. Go2 Walk and Survive are counts over 640 episodes. Online MBD detects, diagnoses, and repairs faults within each episode. Go1 and ANYmal C report walking over 352 episodes per method. On these two robots, the GBM predicts actions directly and shares neither MBD’s gait model nor its input representation.
Fault diagnosis identifies, localizes, and explains failures in engineered and computational systems using sensor data, logs, models, and domain knowledge.