Capstone Project

Problem

• Human–Robot Collaboration (HRC) promises performance gains in domains where humans and robots must act simultaneously, yet most autonomy frameworks treat control as discrete or hierarchical rather than shared

• Existing shared autonomy approaches often lack a principled, continuous formulation, making it difficult to reason about autonomy, trust, and cognitive load in a measurable and repeatable way

Why it matters

• If shared autonomy is to be deployed in safety-critical domains (space operations, surgery, tri-manual assembly), autonomy must be continuous, interpretable, and tunable, not binary or opaque

• Understanding how varying autonomy levels affect human performance, trust, and cognitive load is essential to designing systems that assist rather than overwhelm human operators

• A mathematically grounded autonomy formulation enables systematic experimentation instead of ad-hoc tuning or post-hoc explanations

Approach

• Adopted a Plan–Act interpretation of the Sense–Plan–Act taxonomy, extending prior work that focused solely on execution-level autonomy

• Designed a task environment that forces concurrent collaboration, preventing sequential task decomposition and exposing real coordination effects

• Implemented autonomy as a continuous control blending problem, allowing smooth transitions between human-led and robot-led behaviour

Key Insight

• Reframed acting autonomy as a signal-fusion problem, not a role-assignment problem

• Demonstrated that continuous autonomy can be cleanly expressed as a convex combination of human and robot intent, providing both mathematical rigour and intuitive interpretability

• Showed that autonomy gains emerge not from replacing human control, but from attenuating noise and hesitation in human input while preserving user intent

Contribution

My specific contribution to the project:

• Implemented acting autonomy in software as a convex combination of human and robot control signals. This formulation guaranteed:

• 1. Smooth interpolation between human-dominant and robot-dominant control

  2. Predictable system behaviour across autonomy levels

  3. A direct experimental handle for studying autonomy–trust–performance trade-offs
     
     

Result

• Higher acting autonomy correlated with improved task performance, increased trust, and reduced cognitive load

• Planning autonomy showed similar but weaker effects, suggesting execution-level assistance delivers the strongest immediate benefits

• Participants perceived autonomy changes consistently with the mathematical formulation, supporting the interpretability of the approach

• Despite a limited pilot study size, trends were stable enough to justify further investigation and scaling

• The novelty and stability of the results led to this project being selected for development into a publishable research paper