Touch needs a stronger test.

The sensor is present. Does it help?

The tactile experiment completed 112 MuJoCo episodes: four calibration, twelve teaching and ninety-six held-out episodes. Each held-out condition succeeded in all 32 trials.

The contact condition averaged 225.35 millimetres of displacement, compared with 219.73 millimetres for the frozen condition. But the force-ablated condition also averaged 225.35 millimetres.

Why the ablation matters

A small improvement over frozen behaviour cannot be attributed to tactile information when removing the force input leaves the same result. The appropriate conclusion is that this experiment did not demonstrate a tactile-specific advantage.

The run recorded sensor readings and learned parameters, but those counts do not replace the control comparison. The two seeds reused identical schedules, so they also do not provide independent replication. A stronger task would need to make tactile information causally necessary.