Research Challenges in the CarTel Mobile Sensor System Samuel Madden Associate Professor, MIT
Mar 31, 2015
Research Challenges in the CarTel Mobile Sensor System
Samuel MaddenAssociate Professor, MIT
Wide Area Sensing• Real-world problems:
– Civil infrastructure monitoring– Road-surface conditions– Visual mapping– Commute time optimization
• Wide-area, static sensing– Costly deployment & maintenance
• Observation: some apps do not need high temporal fidelity
• Mobile Sensing– Costly platform?
Our Approach: Opportunistic Mobility
• Take advantage of existing mobility• Example: cellphones w/ sensors
– 1.5 billion phones worldwide– High spatial coverage– High-performance processor
• Cars equipped with sensors– 650 million cars on the road– Abundance of power and space– Have >100 embedded sensors
What system architecture is best suited for mobile, wide-area sensing?
CarTel: A Mobile Sensor Computing System
• Tool to answer questions about spatially diverse data sets– E.g., Collect traffic flow data from every road / issue queries for
route planning
• Core tasks:
1. Collect / process
2. Deliver
3. Visualize / analyze
data from mobile sensors (cars, phones, etc)
Deployment
• Deployed on 9 users’ cars, 27 taxis• 2 boxes per cab
– Master; services for company, drivers, GPS– Slave; experimental box
• Taxi company gets fleet management software, in-car WiFi
• We get data!
• Demo
Coverage Map
Applications & Research
• Route Planning– Under submission
• Pothole Finding– MobiSys 2008
• Managing lossy & noisy trajectories– SIGMOD 2008
• Others – wireless networking (MobiCom 06, 08), carbon footprint, visual mapping, ….
Route Planning
• Match traces to map• Compute Gaussian delay for
each segment– Assume independence
• Minimize 3 metrics– Distance
• Google Maps– Expected delay– Pr(missing time goal)
Max. Probability Planning• Travel time of each edge is a Gaussian
– If indepdendent, travel time of a path is also Gaussian
• Goal: find path with max. probability of reaching destination by deadline
• Unlike standard shortest paths, no suboptimality– If AxCyB is best path from A to B, AxC is not necessarily the best path
from A to C
• Implies cannot use A* or Dijkstra
2
A BC
13Lim et al. “Stochastic Motion Planning and Applications to
Traffic.” Under submission.
Finding Potholes
Classification-based Approach
• Classifier differentiates between several types of anomalies
• Window data, compute features per window
• Variety of features:– Range of X,Y,Z accel– Energy in certain frequency
bands– Car speed– …
See Erikkson et al, MobiSys 2008
FunctionDB
• Challenge: how to store and query all of this data?
• Discrete points don’t work well• Most users don’t actually want raw data!
– Prefer trajectories, fields, fit functions– Idea: support these as first class objects inside the
DBMS
FunctionDB• DBMS that can fit continuous functions to raw
data, query data represented by these functions using SQL
Raw data (temp readings)
Query: Report when temp crosses threshold
SELECT time WHERE temp = thresh
Regression Function temp(t)
Solve equation temp(t) = thresh
time
• Works for any polynomial function
• Supports aggregates (integrals) and joins
• Tricks to deal with intractable queries
• 5-6 x performance gains for common queries on CarTel data
See Thiagarajan and Madden, SIGMOD 2008
temp
Open Problems
• CarTel is a lot of application specific code
• Many SIGMOD papers in building “a declarative framework for X”, where X in {– Signal processing & data management– Personalization– Data cleaning and de-noising– …}
• Focusing on a specific (real) application ensures relevance– Highlights limitations of a database-specific approach
Conclusion
• Research is in capturing, processing, and synthesizing the data– This is what most of us are good at
• This kind of end-to-end deployment isn’t hard– Hardware is $50-$300 / car– 10 cars is sufficient to provide a very interesting data set
• Motes and TinyOS are an interesting novelty, not all there is to sensor networking
• Find an application that excites you and go for it!