Understand this story
On November 20, 2026, UCSD's Hao AI Lab made GPT-6 Astra and Codex robotic agents about three times faster on test tasks by consolidating tool-call round trips, shortening context, and allowing model reasoning and robot movement to run in parallel. The cup task was shortened from 11.1 minutes to 3.4 minutes, while towel folding fell from about 18 minutes to about 6 minutes, and the share of time spent on robot movement rose from 24% to 84%. Each action is still based on observations from about five seconds earlier, and the system takes roughly five seconds to react after the environment changes; there are no other evidence-backed developments at present.
AI-generated from public reports
Latest development · Report
Hao AI Lab Makes GPT-6 Astra and Codex Robotic Agents About 3× Faster at TasksDevelopments
1 confirmed developments
- Source:UCSD Hao AI Lab
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 20
UCSD Hao AI LabFirst-party
Hao AI Lab Makes GPT-6 Astra and Codex Robotic Agents About 3× Faster at Tasks
Hao AI Lab made GPT-6 Astra and Codex agents run about 3× faster on test tasks by consolidating tool-call round trips, shortening context, and allowing model reasoning to run in parallel with robot movement. The cup task was shortened from 11.1 min to 3.4 min, while towel folding was reduced from about 18 min to about 6 min, and the proportion of time spent on robot movement increased from 24% to 84%. Each action is still based on observations from about 5 s earlier, so the response delay after an environmental change is about 5 s.
UCSD Hao AI LabFirst-partySelected
UCSD Hao AI Lab Speeds Up GPT-6 Astra + Codex Robotic Agent by About 3×
By consolidating tool-calling round trips, streamlining context, and allowing the model to think while the robot moves, UCSD Hao AI Lab made the GPT-6 Astra + Codex agent about 3× faster on test tasks. The cup task was shortened from 11.1 min to 3.4 min, while towel folding was reduced from about 18 min to about 6 min, and the share of time spent on robot motion increased from 24% to 84%. Each action is still based on an image from about 5 s earlier, and responding to environmental changes also takes about 5 s.
UCSD Hao AI LabFirst-partySelected
Hao AI Lab Speeds Up GPT-6 Astra + Codex Robotic Agent Tasks by About 3×
By consolidating tool-call round trips, shortening context, and allowing the model to think while the robot moves, Hao AI Lab made the GPT-6 Astra + Codex agent about 3x faster on test tasks. The cup task was reduced from 11.1 min to 3.4 min, while towel folding fell from about 18 min to about 6 min, and the share of time spent on robot movement rose from 24% to 84%. However, each action is still based on visual information from about 5 s earlier, resulting in a response delay of about 5 s after environmental changes. The blog is at https://t.co/BIaiZ2P5ch, and the code is at https://t.co/MvWxzfMXiK.
UCSD Hao AI LabFirst-party
Hao AI Lab Lets Robots Think While Moving, Boosting Task Speed by About 3x in Real-World Tests
Hao AI Lab pipelined the reasoning of GPT-6 Astra + Codex agents with robot motion, making test tasks run about 3x faster: the cup task was shortened from 11.1 min to 3.4 min, while towel folding was reduced from about 18 min to about 6 min. The agent writes actions as Python functions, which the robot executes sequentially in the background while returning results; meanwhile, the model plans or rewrites subsequent steps. The team also consolidated tool-call round trips and converted older images into cropped text notes. Each action is still based on observations from about 5 s earlier, so the response time after an environmental change is approximately 5 s.
UCSD Hao AI LabFirst-partySelected
Hao AI Lab Speeds Up GPT-6 Astra + Codex Robotic Agent by About 3×
By consolidating tool-call round trips, shortening the context, and running model reasoning in parallel with robot motion, Hao AI Lab made the GPT-6 Astra + Codex agent about 3× faster on test tasks. The cup task was shortened from 11.1 min to 3.4 min, while towel folding was reduced from about 18 min to about 6 min, and the share of time spent on robot motion increased from 24% to 84%. Each action is still based on observations from about 5 s earlier, so reacting to environmental changes takes about 5 s.
UCSD Hao AI LabFirst-partySelected
UCSD Hao AI Lab Speeds Up Robotic Agent Tasks by About 3× Through Parallel Reasoning and Motion
UCSD Hao AI Lab has brought a GPT-6 Astra + Codex agent closer to real-time robot control, cutting the cup task from 11.1 min to 3.4 min and towel folding from about 18 min to about 6 min. By consolidating tool-call round trips, compressing context, and allowing the robot to keep moving while the model reasons, the team increased the share of time spent on robot motion from 24% to 84%. Each action is still based on an observation from about 5 s earlier, so reacting to environmental changes also takes about 5 s.
UCSD Hao AI LabFirst-party
UCSD Hao AI Lab Speeds Up GPT-6 Astra and Codex Robotics Agents by About 3×
UCSD Hao AI Lab sped up the tested robotics agents by about 3× by consolidating tool-call round trips, streamlining context, and allowing model reasoning and robot movement to run in parallel. The cup task was shortened from 11.1 min to 3.4 min, while towel folding was reduced from about 18 min to about 6 min, and the proportion of time spent on robot movement increased from 24% to 84%. However, each action is still based on observations from about 5 s earlier, resulting in a response delay of about 5 s after the environment changes.
UCSD Hao AI LabFirst-party
Hao AI Lab Speeds Up GPT-6 Astra Robotic Agents by Consolidating Tool-Call Round Trips
Hao AI Lab consolidated tool-call round trips and enabled robot motion to run in parallel with model reasoning, making the GPT-6 Astra + Codex agent run approximately 3x faster on test tasks. The cup task was reduced from 11.1 min to 3.4 min, towel folding from ~18 min to ~6 min, and the proportion of time spent on robot motion increased from 24% to 84%. The team also made each action return camera imagery and robot state simultaneously, and compressed older images into text notes to shorten the context, although actions were still based on observations from about 5 s earlier.
UCSD Hao AI LabFirst-party
Hao AI Lab Speeds Up GPT-6 Astra + Codex Robotic Agent Tasks by About 3×
By optimizing the GPT-6 Astra + Codex agent, Hao AI Lab cut the cup task from 11.1 min to 3.4 min and towel folding from ~18 min to ~6 min, while increasing the proportion of time the robot spent moving from 24% to 84%. The team consolidated tool-call round trips, compressed older images into text notes, and allowed the robot to move while the model reasoned in parallel, though the response delay after environmental changes remains about 5 s. Blog: https://t.co/BIaiZ2P5ch. Code: https://t.co/MvWxzfMXiK.
AI Trends