Search papers, labs, and topics across Lattice.
This paper introduces PredVLA, a novel language-conditioned predictive-coding policy for robot manipulation that operates with just 0.68 million parameters and without any pretraining on robot data. By leveraging hierarchical generative recurrent dynamics, PredVLA predicts visual features and proprioception, achieving impressive success rates of 86.9% on short-horizon tasks and 75.4% on long-horizon tasks in the LIBERO benchmark. Notably, it outperforms parameter-matched Transformer and LSTM policies by factors of 3.7x and 7.4x, respectively, while providing a clear mechanism for understanding the impact of prediction-error-driven state corrections.
A sub-million-parameter robot manipulation policy outperforms larger models by leveraging predictive coding for real-time state correction.
Large pretrained vision-language-action models dominate modern robot-manipulation benchmarks, but it remains unclear how much model scale is necessary for strong language-conditioned control, or whether fundamentally different control architectures can remain competitive at much smaller parameter budgets. We present PredVLA, a language-conditioned predictive-coding policy with only 0.68 million trainable network parameters and no robot-data pretraining, whose hierarchical generative recurrent dynamics predict visual features and proprioception while observations influence latent state only through online inference from the resulting sensory prediction errors. On LIBERO, PredVLA achieves an 86.9% mean success rate across the three short-horizon suites and 75.4% when the long-horizon suite is included. Under a controlled comparison using the same frozen front end, demonstrations, action decoder, and evaluation protocol, PredVLA achieves 3.7x and 7.4x mean success rates of parameter-matched Transformer and LSTM policies, respectively. The predictive-coding formulation also makes the contribution of observation-driven correction directly measurable: because observations influence the recurrent state only through prediction-error-based latent inference, disabling this inference yields an exact open-loop control condition. Together, these results show that a sub-million-parameter recurrent generative policy can achieve strong performance on modern language-conditioned manipulation benchmarks while providing an explicit mechanism for prediction-error-driven online state correction.
