Sanjiban Choudhury
  sanjiban at cmu dot edu

I am a Postdoctoral fellow at the School of Computer Science and Engineering of University of Washington , where I work with Sidd Srinivasa. I work on applying machine learning techniques to solve problems in motion planning. I recently graduated with a PhD from The Robotics Institute of Carnegie Mellon University, where I was advised by Sebastian Scherer. As part of my thesis work, I developed the motion planning architecture, that integrated learning and planning, for an autonomous full-scale helicopter as part of ONR's AACUS project. During my PhD, I have also worked with Sidd Srinivasa on several fundamental problems in planning and with Drew Bagnell on applying percolation theory to planning.

I spent a lovely summer (2017) at Microsoft Research where I worked with Debadeepta Dey, Ashish Kapoor and Gireeja Ranade on adaptive information gathering via imitation learning. In 2013, I completed my Master’s in Robotics at CMU where I was co-advised by Sanjiv Singh and Sebastian Scherer. I did my Bachelors and Masters in EE at IIT Kharagpur where I started the Kharagpur Robosoccer Group.

I am a Siebel’s Scholar, class of 2013.

CV    /    Google Scholar    /    Bitbucket    /    Github


  • Feb 2018: AACUS is nominated for the Collier Trophy! Previous winners include Apollo 11 and Boeing 747 teams.
  • Feb 2018: I defended my PhD on Adaptive Motion Planning! My thesis and talk are both available.
  • Jan 2018: Our invited paper on Data-driven Planning via Imitation Learning to International Journal of Robotics Research (IJRR) is now available on arXiv.
  • Dec 2017: At NIPS 2017, I presented our work on Near-Optimal Edge Evaluation in Explicit Generalized Binomial Graphs.
  • Dec 2017: Blue sky paper on A Bayesian Active Learning Approach to Adaptive Motion Planning accepted at ISRR 2017.
  • Oct 2017: Our work on Learning Heuristic search via Imitation was one of 18 papers selected for full length track at Conference on Robotic Learning (CoRL) 2017 .
  • Sep 2017: I co-organized a workshop at IROS 2017 on Complex Collaborative Systems. I also gave an invited talk on adaptive motion planning.
  • Jul 2017: I presented our work on Adaptive Information Gathering via Imitation Learning at Robotics Science and Systems (RSS), 2017. Our paper was selected for an invited submission to a special issue of International Journal of Robotics Research (IJRR).
  • Jun 2017: Three papers presented at ICRA 2017



Motion Planning as Bayesian Active Learning

We draw a novel equivalence between motion planning and the Bayesian active learning paradigm of decision region determination (DRD). Given priors on the validity of edges, our goal is to evaluate a sequence of edges that drive uncertainty into a single decision region where a path is valid.

Learning Heuristics via Imitation of Optimal Planners

We explore the problem of learning a heuristic policy that takes as input the state of the search, i.e all decisions undertaken and outcomes observed, and decides which vertex to expand. We explore efficient methods to train such heuristics by imitating optimal planners.


Learning to Gather Information via Imitation

We present a novel data-driven imitation learning framework to efficiently train information gathering policies. The policy imitates a clairvoyant oracle - an oracle that at train time has full knowledge about the world map and can compute maximally informative sensing locations.

Needle in a Haystack: Densification in Planning

We consider the problem of computing shortest paths in a dense motion-planning roadmap. Our key insight is to provide existing path-planning algorithms with a sequence of increasingly dense subgraphs of the roadmap.

A Kite in the Wind: Trajectory Optimization in a Moving Frame

We address the problem of planning long, dynamically feasible, time-optimal trajectories in the presence of wind (which creates a moving reference frame). We present an algorithm, KITE, that elegantly decouples the joint trajectory optimization problem into individual path optimization in a fixed ground frame and a velocity profile optimization in a moving reference frame.


Hybrid Local and Global Search

We present a hybrid technique that integrates the benefits of sampling based optimal planners and local trajectory optimization. Our key insight is that applying local optimization to a subset of edges likely to improve the solution avoids the prohibitive cost of optimizing every edge in a global search. This is made possible by Batch Informed Trees (BIT*), an informed global technique that orders its search by potential solution quality.

Learning to Predict an Ensemble of Planners

Predicting single options in motion planning often leads to scenarios where the prediction suffers a high loss. This is due to the nonsmooth nature of planner performances due to small perturbations in obstacle configurations. We investigate list prediction. Each predictor in a list focusses on increasingly harder problems thus improving worst case performance.

Theoretical Planning Limits using Markov Chains

We examine the problem of motion planning on a resolution constrained lattice for a robot with non-linear dynamics operating in an environment with randomly generated disc shaped obstacles sampled from a homogeneous Poisson process. We use a novel approach that maps the problem to parameters of directed asymmetric hexagonal lattice bond percolation. We map the lattice to an infinite absorbing Markov chain and use results pertaining to its survival to obtain bounds on the parameters.

Optimal Repairing of Vector Fields

This paper presents a framework that integrates vector field based motion planning techniques with an optimal path planner. Our framework uses a vector field as a high level specification of a task and an optimal motion planner (in our case RRT*) as a local, on-line planner that generates paths that follow the vector field, but also consider the new obstacles encountered by the vehicle during the flight.


Adaptive Motion Planning using an Ensemble of Planners

We present an approach that constructs an ensemble of planners to execute in parallel. Our approach optimizes the submodular selection criteria with a greedy approach and lazy evaluation. We seed our selection with learnt priors on planner performance, thus allowing us to solve new applications without evaluating every planner on that application.

Fast Nonlinear Trajectory Optimization using Filtering Techniques

We propose a fast non-linear trajectory optimization by decoupling the workspace optimization from the enforcement of non-linear constraints. We introduce the Dynamics Projection Filter, a nonlinear projection operator based approach that first optimizes a workspace trajectory with reduced constraints and then projects (filters) it to a feasible configuration space trajectory that has a bounded sub-optimality guarantee.

Guaranteed Safe Planning in Partially Known Environments

We present an online algorithm to guarantee the safety of the robot through an emergency maneuver library. The maneuvers in the emergency maneuver library are optimized such that the probability of finding an emergency maneuver that lies in the known obstacle free space is maximized. We prove that the related trajectory set diversity problem is monotonic and sub- modular which enables one to develop an efficient trajectory set generation algorithm with bounded sub-optimality.


Approximate 3D Visibility Graphs

We have designed an algorithm which plans rapidly through free space and is efficiently guided around obstacles. In this paper we present SPARTAN (Sparse Tangential Network) as an approach to create a sparsely connected graph across a tangential surface around obstacles.


Planning Alternate Routes using a Single Sampling Based Planner

We address the problem of autonomously landing a helicopter during an emergency. We designed a planning system to generate alternate routes (AR). This paper presents an algorithm, RRT*-AR, building upon the optimal sampling-based algorithm RRT* to generate AR in realtime.