My research aims to build a unified algorithmic framework where a robot efficiently infers the optimal value function from a bounded set of interactions with both humans and the environment. It ties together insights from motion planning and imitation learning and applies them to robots deployed in the wild.

2013 - 2017


High Performance Flight of Full-Scale Helicopters from Takeoff to Landing

We present the first comprehensive approach to planning safe trajectories for autonomous helicopters from takeoff to landing. We address several problems (1) decoupling the intractable planning problem while retaining some guarantees (2) satisfying complex dynamic constraints which depend on wind and payload (3) real-time planning to avoid obstacles at high speed (4) guaranteeing safety even under software / sensor failure. We validate the approach on 3 different helicopters with 109 hours of autonomous tests and 590 hours of pilot-in-loop test.
Papers:  AHS'14 (best paper),  JFR'19     /   Videos: Clip 1Clip 2Clip 3Clip 4Clip 5

2018 - 2019


Autonomous drone cinematographer

We have designed a selfie drone that can film a target moving in a cluttered environment with almost no prior information. We address several problems for real-time cinematography: (1) vision-based target estimation; (2) 3D signed-distance mapping for occlusion estimation; (3) efficient trajectory optimization for long time-horizon camera motion; and (4) learning-based artistic shot selection. Our system is able to film people, bikes, cars moving outdoors through dirt trails and vegetation.
Papers:  ISER'18IROS'19JFR'19,    /   Videos: Clip 1Clip 2Clip 3

2014 - 2015


Theoretical Planning Limits using Markov Chains

How fast can a drone fly in a forest even if it knew the location of every single tree? We answer this question with the help of percolation theory on random graphs. Specifically, we map the problem to survival of an infinite absorbing Markov chain for which results are known. Hence given the dynamics of any flying system, we can derive a limit on the speed to guarantee safe flight.
Papers:  RSS'15CMU-TR   /   Videos: Clip 1Clip 2

2014 - 2015


Guaranteed Safe Planning in Partially Known Environments

How can we guarantee safety of a high speed robot when it's doesn't know the environment beyond the sensing horizon? We propose a simple idea - only move to a state if it has a emergency maneuver that keeps the robot is known free space. Even in the worst case - a wall appears, planner segfaults, sensor fails - the robot always has a safe trajectory. We formalize the problem of searching for these maneuvers as a set-cover problem which can be solved efficiently. We evaluate this on an actual helicopter by switching off the sensors mid flight.
Papers:  AHS'14ICRA'15   /   Videos: Clip 1Clip 2

2012 - 2014


Autonomous River Exploration

We look at the problem of mapping a river's geometry to help understand the topology and health of an environment. Satellitle imagery is not always possible due to occlusion from vegetation. We develop a micro air vehicle (MAV) that operates beneath the tree line, detects and maps the river, and plans paths around three-dimensional (3D) obstacles (such as overhanging tree branches) to navigate rivers purely with onboard sensing, with no GPS and no prior map. We present fully autonomous flights on riverine environments generating 3D maps over several hundred-meter stretches of tight winding rivers.
Papers:  JFR'15   /   Videos: Clip 1

2011 - 2012


Autonomous Emergency Landing of a Helicopter

When the engines of a helicopter fails, it can still land safely by quick and coordinated control. We address this problem in its entirety: hard time-constraints, challenging terrain, sensor limitations and uncertainty about landing sites. We designed a planning system that deals with all these factors by computing diverse alternate routes. These routes allow pilots to choose a safe landing site as well as act as contingencies in case of unexpected hazards. We evaluate both in simulation and recorded flight data.
Papers:  AHS'13   /   Videos: Clip 1Clip 2