GBB Embodied Interaction AI Institute2026.10 · Research agenda
Research question
Human confirmation should extend beyond a final approval button. When goals are ambiguous, actions are irreversible or the environment changes, people need to understand why the system stopped and how to correct it.
Research method
Design four types of interaction: clarifying intent, confirming before action, interrupting execution and correcting outcomes. Each presents the current goal, key evidence and available choices so people can act without reconstructing the entire system.
Validation approach
Use ambiguous instructions to check whether the system identifies what needs clarification.
Interrupt execution and observe whether its state can be recovered.
Ask participants to change the goal and check whether the workflow still follows the old plan.
Human–AI collaboration starts with clear boundaries for autonomous action. Low-confidence inputs prompt clarification, environmental changes trigger replanning, and human takeover freezes automatic tasks. Before resuming, the system rechecks the environment instead of continuing from an invalid intermediate state.
Experimental design & observations
Stage
What to record
What to check
Intent clarification
Ambiguous instruction, candidate goals
Does the system ask an answerable question?
Interrupt & take over
Action stage, cancellation conditions
Does the system stop promptly and show its state?
Resume execution
New environment snapshot, user confirmation
Is the previous plan revalidated?
Making results comparable
Fix the initial map, endpoints and input version, then change one environmental condition at a time. Record the baseline before introducing obstacles or conflicting inputs. Include successful paths and failure cases. Compare strategies using path length, explored cells and execution traces.
Reproducibility checklist
Input and map snapshots: record scene versions, task conditions and random seeds.
Strategy configuration: record algorithms, constraints, timeouts and human intervention points.
Execution traces: retain state changes, action sequences, failure causes and recovery steps.
Comparison notes: document variable changes, observations and follow-up validation.
