roboticsopinionbearishSimply increasing number of viewpoints is insufficient for spatial generalization in VLA modelsComputer Vision27 Jul 2026http://arxiv.org/abs/2607.02322v1