TBNAI
Conditional Trajectory Peaks: Single-Pass Multimodal Policies over Action Chunks
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 04/10/2026, 20:00:00 UTC
Briefing created: 27/10/2026, 07:33:53 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 07/10/2026, 11:18:24 UTC.
Briefings
A single-pass multimodal policy framework named Conditional Trajectory Peaks (CTP) was developed to address limitations in imitation learning by predicting complete action-chunks, probability masses, and trajectory scales simultaneously. The system achieves a coverage score of 91.40% on Push-T and maintains high success rates across multiple benchmarks, including 100% on D3IL Avoiding and 97.25% on LIBERO. Real-world dual-arm experiments showed CTP successfully preserved placement modes in a two-plate task for all 50 trials. The model reduced inference latency from 218.24 ms to 75.80 ms while maintaining comparable performance to baseline policies like pi_0.5.
Key points
- CTP is a single-pass policy framework that jointly predicts complete action-chunk candidates, probability masses, and trajectory scales.
- CTP achieved a coverage score of 91.40% on Push-T and maintained 100% success rates on D3IL Avoiding.
Context
Imitation learning requires diverse executable futures under the same observation and consistent behavior across replanning cycles. This research presents CTP, a single-pass policy framework that addresses these requirements by predicting complete action-chunks, probability masses, and trajectory scales simultaneously.
Limitations
Evidence is limited to specific benchmarks and real-world dual-arm experiments. The study does not report results on other tasks or environments, nor does it provide a comprehensive evaluation of the model's performance across all possible scenarios.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 27/10/2026, 07:33:53 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihmholuek7rnn6jbq5nmalruhtltqksipnpqifhsxgsujiiljv3mi