FLEX Teaches Robots to Generalize From Human Corrections
FLEX Teaches Robots to Generalize From Human Corrections
MIT and NYU researchers developed FLEX, a system that uses conversational corrections to revise environmental features guiding robot movements. The Flexible Feature Editing Pipeline separates feature specification from reward optimization, using an ADD, GENERALIZE, and MASK process to handle incomplete or over-specified representations. Tested in simulated navigation grids and Franka tabletop tasks, FLEX achieved the highest average normalized reward (0.99) compared to baselines (0.69–0.83). Human ratings showed superior intent understanding (6.3/7.0 vs 5.0 and 4.5). The system is currently limited to quasi-static manipulation of rigid objects in simulation, with physical closed-loop operation pending further development.
Source: Robotics — Robot learning news discovery · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S7
Reported numbers
normalized reward
0.99
View original evidence
orst-case trajectory. The paper reports that FLEX achieved the highest average normalized reward in all seven navigation scenarios. In the scenario requiring all three operations, FLEX averaged 0.99, compared with 0.69 for ALGAE, 0.83 for MFE and 0.78 for the unscaffolded baseline. In the sequential arm simulations, the authors report that FLEX generally learned the target reward after one or two corrections and main
Open source S9normalized reward
0.69
View original evidence
orst-case trajectory. The paper reports that FLEX achieved the highest average normalized reward in all seven navigation scenarios. In the scenario requiring all three operations, FLEX averaged 0.99, compared with 0.69 for ALGAE, 0.83 for MFE and 0.78 for the unscaffolded baseline. In the sequential arm simulations, the authors report that FLEX generally learned the target reward after one or two corrections and main
Open source S9normalized reward
0.83
View original evidence
orst-case trajectory. The paper reports that FLEX achieved the highest average normalized reward in all seven navigation scenarios. In the scenario requiring all three operations, FLEX averaged 0.99, compared with 0.69 for ALGAE, 0.83 for MFE and 0.78 for the unscaffolded baseline. In the sequential arm simulations, the authors report that FLEX generally learned the target reward after one or two corrections and main
Open source S9normalized reward
0.78
View original evidence
orst-case trajectory. The paper reports that FLEX achieved the highest average normalized reward in all seven navigation scenarios. In the scenario requiring all three operations, FLEX averaged 0.99, compared with 0.69 for ALGAE, 0.83 for MFE and 0.78 for the unscaffolded baseline. In the sequential arm simulations, the authors report that FLEX generally learned the target reward after one or two corrections and main
Open source S9intent understanding score
6.3
View original evidence
all five questionnaire dimensions. For intent understanding on a seven-point scale, the averages were 6.3 for FLEX, 5.0 for ALGAE and 4.5 for MFE. FLEX’s differences from both baselines were statistically significant after adjustment for multiple comparisons. Differences in ease of use and collaborativity were not statistically significant. Evidence Limits and Development Status The human study did not run closed-lo
Open source S11intent understanding score
5
View original evidence
all five questionnaire dimensions. For intent understanding on a seven-point scale, the averages were 6.3 for FLEX, 5.0 for ALGAE and 4.5 for MFE. FLEX’s differences from both baselines were statistically significant after adjustment for multiple comparisons. Differences in ease of use and collaborativity were not statistically significant. Evidence Limits and Development Status The human study did not run closed-lo
Open source S11intent understanding score
4.5
View original evidence
all five questionnaire dimensions. For intent understanding on a seven-point scale, the averages were 6.3 for FLEX, 5.0 for ALGAE and 4.5 for MFE. FLEX’s differences from both baselines were statistically significant after adjustment for multiple comparisons. Differences in ease of use and collaborativity were not statistically significant. Evidence Limits and Development Status The human study did not run closed-lo
Open source S11
- paper: not_reported
- code: not_reported
- dataset: not_reported
- weights: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
at up from a sentence or two instead of asking you to spell out every rule — and that’s the difference between a robot people put up with and one they actually want in their home.” Simulation Results and Human Ratings The researchers tested FLEX in 9-by-9 navigation grids and simulated seven-degree-of-freedom Franka tabletop tasks involving coffee, a lit candle and a bowling ball. The arm scenarios used five-stage co
Open source S7
orst-case trajectory. The paper reports that FLEX achieved the highest average normalized reward in all seven navigation scenarios. In the scenario requiring all three operations, FLEX averaged 0.99, compared with 0.69 for ALGAE, 0.83 for MFE and 0.78 for the unscaffolded baseline. In the sequential arm simulations, the authors report that FLEX generally learned the target reward after one or two corrections and main
Open source S9
all five questionnaire dimensions. For intent understanding on a seven-point scale, the averages were 6.3 for FLEX, 5.0 for ALGAE and 4.5 for MFE. FLEX’s differences from both baselines were statistically significant after adjustment for multiple comparisons. Differences in ease of use and collaborativity were not statistically significant. Evidence Limits and Development Status The human study did not run closed-lo
Open source S11
tained performance on earlier and later configurations. The comparison methods required more corrections and degraded as irrelevant features accumulated or their representations overfit individual stages. A separate in-person study involved 12 participants, each providing typed corrections across three five-stage blocks using FLEX, ALGAE and MFE. The researchers report that FLEX received the highest average rating on
Open source S10
Source：Robotics — Robot learning news discovery · unite.ai